PaddleDetection目标检测任务简介和概述
2024-04-20 20:08:42
本篇博客不计入作业
目标检测任务简介
目标检测的主要目的是让计算机可以自动识别图片或者视频帧中所有目标的类别，并在该目标周围绘制边界框，标示出每个目标的位置，如下图所示。
图（a）是图像分类任务，只需识别出这是一张斑马的图片。 图（b）是目标检测任务，不仅要识别出这是一张斑马的图片，还要标出图中斑马的位置。
目标检测是计算机视觉的主要方向之一，同时也是实例分割，人脸检测，关键点检测等视觉任务的基础，目标检测应用场景广泛，下面列举了一些常见的应用场景。
目标检测任务的挑战
对于目标检测任务，主要存在以下挑战：
- 数据有限(Limited data)
目标检测的样本标注比图像分类要难很多，所以很难有像ImageNet这样的超大数据集用于训练，而数据量对模型非常重要，因此在目标检测中数据增强是十分重要的。 - 定位/识别任务双优先级(Dual priority)
目标检测涉及定位和识别两个子任务，高效地在两个子任务中都取得很好的效果，不同算法有不同的效果，不同场景需要不同的侧重。 - 多尺度(Multiple spatial scales and aspect ratios)
目标检测中存在多尺度的情况，比如狗和小黄鸭，尺度相差很大，框的长宽比也有差异。 - 类别不均衡(Class imbalance)
不同类别样本数差异大的问题，两条腿的人到处都是，四条腿的熊猫只能去动物园找到，类别的不均衡也会加大目标检测任务的难度。 - 实时检测需求(Speed for real-time detection)
不少应用场景对目标检测的实时性要求很高。
PaddleDetection简介
PaddleDetection是飞桨推出的端到端目标检测开发套件，旨在帮助开发者更快更好地完成检测模型的训练、精度速度优化到部署全流程。PaddleDetection以模块化的设计实现了多种主流目标检测算法，并且提供了丰富的数据增强、网络组件、损失函数等模块，集成了模型压缩和跨平台高性能部署能力。目前基于PaddleDetection已经完成落地的项目涉及工业质检、遥感图像检测、无人巡检等多个领域。
PaddleDetection具备以下四大优势：
- 模型丰富：PaddleDetection提供了丰富的模型，包含目标检测、实例分割、人脸检测等100+个预训练模型，10+算法，持续推出针对服务器端和移动端、嵌入式等的增强模型，涵盖多种数据集竞赛冠军方案，并提供适合云端/边缘端设备部署的检测方案。
- 高灵活度：PaddleDetection通过模块化设计来解耦各个组件，模型网络结构和数据处理均可定制，基于配置文件可以轻松地搭建各种检测模型。
- 易部署:PaddleDetection的模型中使用的核心算子均通过C++或CUDA实现，提供跨平台推理引擎，实现了模型训练到部署的无缝衔接，同时内置集成了一键式压缩和部署能力。
- 高性能：基于飞桨框架的高性能内核，实现了模型的高效训练和推理。例如，YOLOv3的训练速度快于其他框架，在Tesla V100 16GB环境下，Mask-RCNN(ResNet50)可以单卡Batch Size可以达到4 (甚至到5)。
PaddleDetection模型库概览
PaddleDetection模块式地提供YOLOv3，EfficientDet等10余种目标检测算法、ResNet-vd，MobileNetV3等10余种backbone，以及sync batch norm, IoU Loss、可变性卷积等多个扩展模块，这些模块可以定制化地自由组合，灵活配置；同时预置提供100余种训练好的检测模型。
服务器端增强模型
PaddleDetection不仅复现了常见的目标检测模型，还对模型的进行了图像增强、骨干网络优化、DropBlock，IoU Loss IoUAware等一系列深度优化。同时，内置集成模型压缩能力，提供了一键式剪裁，蒸馏，量化的脚本，大大提高模型精度和速度，并减小模型体积。
单阶段目标检测模型YOLOv3
精度方面，原论文作者开源的YOLOv3模型在COCO数据集上精度是33.0，PaddleDetectio将其提高到43.6（提高10.6个绝对百分点）；推理速度方面，领先于同类产品40%左右。两阶段目标检测模型FasterRCNN-ResNet50vd-FPN
精度方面，通过减小NMS的proposal，和减小FPN的通道数，虽然精度略有降低，但是推理速度大幅提升；通过PaddleClas里的SSLD蒸馏的预训练、CIoU Loss、自动图像增强等优化方法，将精度优化到41.6（提升5.2个绝对百分点）。推理速度方面，提升了38%，在Tesla V100上的预测速度达到60FPS。模型压缩
在YOLOv3系列模型上，通过一键式剪裁+蒸馏的方案，YOLOv3_MobileNetV1剪裁了近70%的计算量，在精度基本无损或略有提升的情况，模型在高通855芯片上加速2.3倍，GPU上也有60%的加速；YOLOv3-ResNet50vd-DCN剪裁模型，精度提升了0.6，GPU上加速20%。同时，对应压缩后的模型、压缩脚本和操作方法均可以在Github上获取。
端侧轻量化增强模型
对于移动端，嵌入式这些端侧的设备，一般算力比较低，内存较小，所以对运算量和模型大小要求比较高，同时端侧的设备种类也比较多，有早起的低端CPU，也有近年来推出的骁龙855，麒零980等性能较好的CPU，还有更优一些嵌入式的GPU，NPU等。PaddleDetection针对这种情况，推出了三组模型：
- 对于低算力的设备，推出SSDLite及其量化模型，通过模型增强，SSDLite在骁龙845芯片上推理时延达到41ms。
- 对于需要兼顾速度和精度的应用场景，推出YOLOv3剪裁+蒸馏的压缩模型，在100ms左右推理时延上能在COCO数据集上达到25左右的精度。
- 对于算力比较高的设备，推出Cascade Faster RCNN模型，最高能在COCO数据集上达到30.2的精度。
通过这三组增强模型，可以根据自己的设备算力来选择合适的端上模型，下图是三组模型的增强效果。
低端算力：SSDLite模型 + 量化
中端算力：兼顾速度和精度的YOLOv3压缩模型
高端算力：高精度的Cascade Faster RCNN模型
PaddleDetection配置文件
如前文所述，PaddleDetection是飞桨推出的端到端目标检测开发套件，旨在帮助开发者更快更好地完成检测模型的训练、精度速度优化到部署全流程。支持现有的RCNN、SSD、YOLO等系列模型、支持 ResNet、ResNet-VD、ResNeXt、ResNeXt-VD、SENet、MobileNet、DarkNet等主干网络。而快速完成不同模型的训练体验，就是通过配置文件的切换。
亮点：一键式训练/推理/部署
PaddleDetection提供所有模型、算法、扩展模块的配置脚本，同时提供一键式训练、评估、推理、模型导出等脚本，从而帮助开发者完成不同模型的快速体验。
亮点：一键式模型增强
PaddleDetection以组件式形式提供模型的所有优化方法，只需要修改对应配置文件的参数配置，就能一键式模型增强。
PaddleDetection配置文件目录
PaddleDetection├── configs # 提供所有模型、算法、扩展模块的配置脚本├── contrib├── dataset├── demo├── deploy # 提供C++端和python端使用飞桨预测库进行推理部署的详细指导├── docs├── ppdet├── slim├── tools # 一键式训练、评估、推理、模型导出`train.py` `eval.py` `infer.py` `export_model.py`├── .gitignore├── .pre-commit-config.yaml├── .style.yapf├── .travis.yml├── LICENSE├── README.md├── README_en.md├── requirements.txt # 列出了PaddleDetection的所有依赖库常用命令
在命令提示符下输入命令时，一定避免输入多余的空格。比如，weights=后面出现了一个多的空格，导致模型文件读不到。
-c: 指定配置文件进行一键式训练
--eval：训练中使用评估
-o use_gpu=False：使用CPU训练。不建议在CPU下训练模型，若只能使用CPU运行PaddleDetection，需指定。
这里以faster_rcnn_r50_1x.yml举例，展示配置文件的使用方法。
# 设置PYTHONPATH路径
export PYTHONPATH=$PYTHONPATH:.
# GPU训练 支持单卡，多卡训练，通过CUDA_VISIBLE_DEVICES指定卡号
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
python tools/train.py -c configs/faster_rcnn_r50_1x.yml
# GPU评估
export CUDA_VISIBLE_DEVICES=0
python tools/eval.py -c configs/faster_rcnn_r50_1x.yml
# 预测
python tools/infer.py -c configs/faster_rcnn_r50_1x.yml --infer_img=demo/000000570688.jpg常用参数设置
训练配置文件需要着重关注的参数如下：（需要根据自己的不同数据集进行训练）
- 最大迭代步数：max_iters
- 预训练模型的来源：pretrain_weights
- 数据路径：dataset_dir
- Batch_size的大小：batch_size
- 数据集类别：num_classes
- 学习率：LearningRate.base_lr
下文“十分钟入门，玩转视频的目标检测”，使用PaddleDetection提供的一键式脚本，带你快速完成视频流的目标检测任务。
十分钟入门，玩转视频的目标检测
以YOLOv3_darknet模型为例，介绍如何使用PaddleDetection进行目标检测模型的训练和部署，以及对视频流进行目标检测。
注意事项
请使用AI Studio的“高级版”运行本节相关程序，如下图所示。
模型训练期间，需要确保此网页不关闭、电脑屏幕未锁定，否则训练将中断。
除“数据集”和“文件夹 > Work”目录外，其他输出文件在重启环境时将被清空。如果希望本次实验任务用到的文件在下次进入环境时仍然存在，可以选择保存到这两个地方。
环境准备
PaddleDetection下载
可以通过如下git clone命令下载PaddleDetection目标检测库，由于在AIStudio上通过git clone下载比较慢，本项目在work目录下提供下载好的PaddleDetection压缩包，也可直接解压使用。
%cd data
# !git clone https://github.com/PaddlePaddle/PaddleDetection # 下载PaddleDetection目标检测库
#! tar xf ../work/PaddleDetection.tar -C . # 解压work目录下已下载好的PaddleDetection压缩包
!unzip -o /home/aistudio/work/PaddleDetection.zip安装PaddleDetection依赖库
通过如下方式安装PaddleDetection依赖，并设置环境变量。
%cd PaddleDetection
!pip install -r requirements.txt # requirements.txt列出了PaddleDetection的所有依赖库
!pip install Cython pycocotools
%env PYTHONPATH=.:$PYTHONPATH
%env CUDA_VISIBLE_DEVICES=0数据准备
PaddleDetection提供了COCO、VOC等数据集下载脚本如dataset/coco/download.py，通过此脚本可自动下载对应数据集。但由于COCO数据集下载比较耗时，本项目已上传COCO2017数据集，可直接解压获取。
COCO的
train数据集较大，解压比较耗时，本项目仅使用val数据集。
# ! python dataset/coco/download_coco.py # coco数据集下载脚本
! unzip -q /home/aistudio/data/data7122/annotations_trainval2017.zip -d dataset/coco # 解压数据集
! unzip -q /home/aistudio/data/data7122/val2017.zip -d dataset/coco
# ! unzip /home/aistudio/data/data7122/train2017.zip -d dataset/coco一键式模型训练
本项目选取yolov3_darknet.yml作为本项目的训练配置文件。该配置文件选取的是YOLOv3_darknet模型，骨干网络为DarkNet,yolo_head为YOLOv3Head。
本项目演示使用COCO的val数据集训练，对配置文件做如下修改（使用train数据集不需要做以下修改，可以直接一键式启动训练），work目录下提供了一个修改好的配置文件，也可在configs目录（即data/PaddleDetection/configs目录）下按照如下方式自行修改yolov3_darknet.yml和yolov3_reader.yml。
1.本次项目将训练轮数max_iters减小到50000, 减少PiecewiseDecay的milestones到[40000, 45000]。
2.修改TrainReader.dataset的anno_path和image_dir为val数据集路径, mixup_epoch修改为0。
3.由于使用单卡训练，将LearningRate.base_lr减少8倍到0.000125。
yolov3_darknet.yml文件配置好之后，就可以进行一键式模型训练。
常见的训练命令：
-c configs/yolov3_darknet.yml：用来指定模型配置文件 –use_tb：是否使用tb-paddle记录数据，进而在TensorBoard中显示，默认值是False –tb_log_dir：指定 tb-paddle 记录数据的存储路径 –eval：是否边训练边测试，在训练中交替执行评估, 评估在每个snapshot_iter时开始。每次评估后还会评出最佳mAP模型保存到best_model文件夹下，建议训练时候使用该参数，可以使得完成训练后快速找到最好的模型。
此处仅展示一键式训练，训练完成比较耗时(4小时左右)。如果想压缩训练时间，可以在
yolov3_darknet.yml中将训练轮数max_iters调小，后续使用PaddleDetection发布的权重（yolov3_darknet模型的训练权重默认存储于output/yolov3_darknet目录）进行推理。
! python tools/train.py -c ../../work/yolov3_darknet.yml --use_vdl=True --vdl_log_dir=/home/aistudio/log --eval # 启动模型训练2020-07-31 15:37:31,475-INFO: If regularizer of a Parameter has been set by 'fluid.ParamAttr' or 'fluid.WeightNormParamAttr' already. The Regularization[L2Decay, regularization_coeff=0.000500] in Optimizer will not take effect, and it will only be applied to other Parameters! 2020-07-31 15:37:32,678-INFO: font search path ['/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/matplotlib/mpl-data/fonts/ttf', '/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/matplotlib/mpl-data/fonts/afm', '/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/matplotlib/mpl-data/fonts/pdfcorefonts'] 2020-07-31 15:37:34,002-INFO: generated new fontManager loading annotations into memory... Done (t=0.64s) creating index... index created! 2020-07-31 15:37:35,710-INFO: places would be ommited when DataLoader is not iterable W0731 15:37:35.749830 206 device_context.cc:252] Please NOTE: device: 0, CUDA Capability: 70, Driver API Version: 10.1, Runtime API Version: 9.0 W0731 15:37:35.753389 206 device_context.cc:260] device: 0, cuDNN Version: 7.3. 2020-07-31 15:37:38,410-INFO: Downloading DarkNet53_pretrained.tar from https://paddle-imagenet-models-name.bj.bcebos.com/DarkNet53_pretrained.tar 100%|████████████████████████████████| 158940/158940 [00:08<00:00, 18874.15KB/s] 2020-07-31 15:37:46,919-INFO: Decompressing /home/aistudio/.cache/paddle/weights/DarkNet53_pretrained.tar... 2020-07-31 15:37:47,148-WARNING: /home/aistudio/.cache/paddle/weights/DarkNet53_pretrained.pdparams not found, try to load model file saved with [ save_params, save_persistables, save_vars ] loading annotations into memory... Done (t=0.72s) creating index... index created! 2020-07-31 15:37:54,357-INFO: places would be ommited when DataLoader is not iterable 2020-07-31 15:37:57,222-INFO: iter: 0, lr: 0.000000, 'loss': '4349.791016', time: 0.006, eta: 0:04:50 2020-07-31 15:38:07,677-INFO: iter: 20, lr: 0.000001, 'loss': '7997.978516', time: 0.654, eta: 9:04:49 2020-07-31 15:38:13,518-INFO: iter: 40, lr: 0.000001, 'loss': '1091.667236', time: 0.292, eta: 4:03:11 2020-07-31 15:38:19,337-INFO: iter: 60, lr: 0.000002, 'loss': '585.337524', time: 0.292, eta: 4:03:01 2020-07-31 15:38:25,974-INFO: iter: 80, lr: 0.000003, 'loss': '506.772156', time: 0.325, eta: 4:30:45 2020-07-31 15:38:33,924-INFO: iter: 100, lr: 0.000003, 'loss': '478.433105', time: 0.391, eta: 5:24:47 2020-07-31 15:38:39,475-INFO: iter: 120, lr: 0.000004, 'loss': '409.666504', time: 0.290, eta: 4:01:14 2020-07-31 15:38:45,500-INFO: iter: 140, lr: 0.000004, 'loss': '385.746246', time: 0.304, eta: 4:12:49 2020-07-31 15:38:51,134-INFO: iter: 160, lr: 0.000005, 'loss': '446.195587', time: 0.265, eta: 3:40:03 2020-07-31 15:38:57,548-INFO: iter: 180, lr: 0.000006, 'loss': '378.182983', time: 0.334, eta: 4:37:18 2020-07-31 15:39:02,861-INFO: iter: 200, lr: 0.000006, 'loss': '425.748291', time: 0.267, eta: 3:41:57 2020-07-31 15:39:09,404-INFO: iter: 220, lr: 0.000007, 'loss': '370.520874', time: 0.314, eta: 4:20:39 2020-07-31 15:39:15,689-INFO: iter: 240, lr: 0.000008, 'loss': '355.822144', time: 0.321, eta: 4:26:29 2020-07-31 15:39:21,506-INFO: iter: 260, lr: 0.000008, 'loss': '354.289124', time: 0.294, eta: 4:03:18 2020-07-31 15:39:27,545-INFO: iter: 280, lr: 0.000009, 'loss': '320.562805', time: 0.293, eta: 4:02:37 2020-07-31 15:39:33,780-INFO: iter: 300, lr: 0.000009, 'loss': '311.270996', time: 0.320, eta: 4:25:05 2020-07-31 15:39:39,910-INFO: iter: 320, lr: 0.000010, 'loss': '267.223694', time: 0.311, eta: 4:17:37 2020-07-31 15:39:46,555-INFO: iter: 340, lr: 0.000011, 'loss': '249.246765', time: 0.332, eta: 4:34:33 2020-07-31 15:39:52,995-INFO: iter: 360, lr: 0.000011, 'loss': '243.238327', time: 0.307, eta: 4:13:37 2020-07-31 15:39:59,105-INFO: iter: 380, lr: 0.000012, 'loss': '210.381363', time: 0.317, eta: 4:22:29 2020-07-31 15:40:04,653-INFO: iter: 400, lr: 0.000013, 'loss': '203.033066', time: 0.284, eta: 3:54:28 2020-07-31 15:40:11,069-INFO: iter: 420, lr: 0.000013, 'loss': '185.442490', time: 0.306, eta: 4:12:39 2020-07-31 15:40:17,031-INFO: iter: 440, lr: 0.000014, 'loss': '216.101425', time: 0.308, eta: 4:14:13 2020-07-31 15:40:23,471-INFO: iter: 460, lr: 0.000014, 'loss': '157.271713', time: 0.326, eta: 4:29:32 2020-07-31 15:40:30,143-INFO: iter: 480, lr: 0.000015, 'loss': '165.912811', time: 0.326, eta: 4:29:18 2020-07-31 15:40:37,252-INFO: iter: 500, lr: 0.000016, 'loss': '200.069427', time: 0.345, eta: 4:44:36 2020-07-31 15:40:43,732-INFO: iter: 520, lr: 0.000016, 'loss': '163.968475', time: 0.328, eta: 4:30:44 2020-07-31 15:40:49,928-INFO: iter: 540, lr: 0.000017, 'loss': '158.648056', time: 0.322, eta: 4:25:06 2020-07-31 15:40:56,070-INFO: iter: 560, lr: 0.000018, 'loss': '155.449524', time: 0.300, eta: 4:07:02 2020-07-31 15:41:02,123-INFO: iter: 580, lr: 0.000018, 'loss': '149.289307', time: 0.300, eta: 4:06:54 2020-07-31 15:41:07,229-INFO: iter: 600, lr: 0.000019, 'loss': '153.428802', time: 0.259, eta: 3:33:21 2020-07-31 15:41:15,485-INFO: iter: 620, lr: 0.000019, 'loss': '141.784515', time: 0.422, eta: 5:47:05 2020-07-31 15:41:21,090-INFO: iter: 640, lr: 0.000020, 'loss': '114.264877', time: 0.280, eta: 3:50:34 2020-07-31 15:41:27,319-INFO: iter: 660, lr: 0.000021, 'loss': '145.880859', time: 0.309, eta: 4:14:12 2020-07-31 15:41:34,049-INFO: iter: 680, lr: 0.000021, 'loss': '144.587112', time: 0.321, eta: 4:24:06 2020-07-31 15:41:40,005-INFO: iter: 700, lr: 0.000022, 'loss': '114.052292', time: 0.314, eta: 4:17:52 2020-07-31 15:41:46,828-INFO: iter: 720, lr: 0.000023, 'loss': '112.328705', time: 0.326, eta: 4:27:55 2020-07-31 15:41:53,791-INFO: iter: 740, lr: 0.000023, 'loss': '129.093109', time: 0.359, eta: 4:54:41 2020-07-31 15:41:58,980-INFO: iter: 760, lr: 0.000024, 'loss': '116.874458', time: 0.252, eta: 3:26:39 2020-07-31 15:42:05,323-INFO: iter: 780, lr: 0.000024, 'loss': '123.895432', time: 0.314, eta: 4:17:19 2020-07-31 15:42:10,919-INFO: iter: 800, lr: 0.000025, 'loss': '120.467987', time: 0.292, eta: 3:59:39 2020-07-31 15:42:17,120-INFO: iter: 820, lr: 0.000026, 'loss': '125.974319', time: 0.301, eta: 4:06:49 2020-07-31 15:42:22,653-INFO: iter: 840, lr: 0.000026, 'loss': '112.528641', time: 0.288, eta: 3:56:21 2020-07-31 15:42:28,742-INFO: iter: 860, lr: 0.000027, 'loss': '118.941055', time: 0.303, eta: 4:08:01 2020-07-31 15:42:35,879-INFO: iter: 880, lr: 0.000028, 'loss': '109.778778', time: 0.342, eta: 4:40:05 2020-07-31 15:42:42,483-INFO: iter: 900, lr: 0.000028, 'loss': '92.892197', time: 0.342, eta: 4:39:33 2020-07-31 15:42:48,342-INFO: iter: 920, lr: 0.000029, 'loss': '128.099503', time: 0.294, eta: 4:00:07 2020-07-31 15:42:54,298-INFO: iter: 940, lr: 0.000029, 'loss': '127.058044', time: 0.301, eta: 4:06:17 2020-07-31 15:43:00,445-INFO: iter: 960, lr: 0.000030, 'loss': '117.944519', time: 0.302, eta: 4:07:02 2020-07-31 15:43:06,757-INFO: iter: 980, lr: 0.000031, 'loss': '118.212769', time: 0.309, eta: 4:12:11 2020-07-31 15:43:12,646-INFO: iter: 1000, lr: 0.000031, 'loss': '124.516418', time: 0.303, eta: 4:07:47 2020-07-31 15:43:18,317-INFO: iter: 1020, lr: 0.000032, 'loss': '94.399078', time: 0.284, eta: 3:51:30 2020-07-31 15:43:24,336-INFO: iter: 1040, lr: 0.000033, 'loss': '113.629181', time: 0.304, eta: 4:07:47 2020-07-31 15:43:30,204-INFO: iter: 1060, lr: 0.000033, 'loss': '110.786438', time: 0.289, eta: 3:55:39 2020-07-31 15:43:35,741-INFO: iter: 1080, lr: 0.000034, 'loss': '85.554443', time: 0.283, eta: 3:50:31 2020-07-31 15:43:41,941-INFO: iter: 1100, lr: 0.000034, 'loss': '111.587654', time: 0.306, eta: 4:09:14 2020-07-31 15:43:47,569-INFO: iter: 1120, lr: 0.000035, 'loss': '96.128357', time: 0.287, eta: 3:53:29 2020-07-31 15:43:53,378-INFO: iter: 1140, lr: 0.000036, 'loss': '104.191391', time: 0.285, eta: 3:51:52 2020-07-31 15:43:59,129-INFO: iter: 1160, lr: 0.000036, 'loss': '115.461334', time: 0.279, eta: 3:46:53 2020-07-31 15:44:04,937-INFO: iter: 1180, lr: 0.000037, 'loss': '100.543701', time: 0.300, eta: 4:04:02 2020-07-31 15:44:11,719-INFO: iter: 1200, lr: 0.000038, 'loss': '125.268829', time: 0.335, eta: 4:32:50 2020-07-31 15:44:18,316-INFO: iter: 1220, lr: 0.000038, 'loss': '113.588531', time: 0.322, eta: 4:21:37 2020-07-31 15:44:27,096-INFO: iter: 1240, lr: 0.000039, 'loss': '102.357216', time: 0.441, eta: 5:58:25 2020-07-31 15:44:33,071-INFO: iter: 1260, lr: 0.000039, 'loss': '106.932518', time: 0.304, eta: 4:07:00 2020-07-31 15:44:39,196-INFO: iter: 1280, lr: 0.000040, 'loss': '111.148056', time: 0.310, eta: 4:11:50 2020-07-31 15:44:45,698-INFO: iter: 1300, lr: 0.000041, 'loss': '96.706154', time: 0.313, eta: 4:14:15 2020-07-31 15:44:52,012-INFO: iter: 1320, lr: 0.000041, 'loss': '101.807487', time: 0.320, eta: 4:19:39 2020-07-31 15:44:58,012-INFO: iter: 1340, lr: 0.000042, 'loss': '122.297989', time: 0.307, eta: 4:09:01 2020-07-31 15:45:04,024-INFO: iter: 1360, lr: 0.000043, 'loss': '103.514076', time: 0.303, eta: 4:05:50 2020-07-31 15:45:10,722-INFO: iter: 1380, lr: 0.000043, 'loss': '85.650970', time: 0.322, eta: 4:20:51 2020-07-31 15:45:16,241-INFO: iter: 1400, lr: 0.000044, 'loss': '98.000046', time: 0.288, eta: 3:53:05 2020-07-31 15:45:22,098-INFO: iter: 1420, lr: 0.000044, 'loss': '104.807053', time: 0.297, eta: 4:00:23 2020-07-31 15:45:28,182-INFO: iter: 1440, lr: 0.000045, 'loss': '108.437813', time: 0.300, eta: 4:03:04 2020-07-31 15:45:34,119-INFO: iter: 1460, lr: 0.000046, 'loss': '118.776573', time: 0.300, eta: 4:02:22 2020-07-31 15:45:40,292-INFO: iter: 1480, lr: 0.000046, 'loss': '93.497322', time: 0.304, eta: 4:05:34 2020-07-31 15:45:46,240-INFO: iter: 1500, lr: 0.000047, 'loss': '96.293488', time: 0.303, eta: 4:04:39 2020-07-31 15:45:52,821-INFO: iter: 1520, lr: 0.000048, 'loss': '106.420120', time: 0.317, eta: 4:16:27 2020-07-31 15:45:58,357-INFO: iter: 1540, lr: 0.000048, 'loss': '99.813301', time: 0.287, eta: 3:52:06 2020-07-31 15:46:04,764-INFO: iter: 1560, lr: 0.000049, 'loss': '104.515434', time: 0.316, eta: 4:14:47 2020-07-31 15:46:11,383-INFO: iter: 1580, lr: 0.000049, 'loss': '86.635101', time: 0.338, eta: 4:32:47 2020-07-31 15:46:17,482-INFO: iter: 1600, lr: 0.000050, 'loss': '97.006958', time: 0.292, eta: 3:55:45 2020-07-31 15:46:23,015-INFO: iter: 1620, lr: 0.000051, 'loss': '93.530777', time: 0.287, eta: 3:51:10 2020-07-31 15:46:29,059-INFO: iter: 1640, lr: 0.000051, 'loss': '93.594009', time: 0.294, eta: 3:56:46 2020-07-31 15:46:35,713-INFO: iter: 1660, lr: 0.000052, 'loss': '99.072800', time: 0.341, eta: 4:34:52 2020-07-31 15:46:41,769-INFO: iter: 1680, lr: 0.000053, 'loss': '75.845039', time: 0.297, eta: 3:58:53 2020-07-31 15:46:47,867-INFO: iter: 1700, lr: 0.000053, 'loss': '98.739716', time: 0.301, eta: 4:02:05 2020-07-31 15:46:54,506-INFO: iter: 1720, lr: 0.000054, 'loss': '101.547394', time: 0.329, eta: 4:24:25 2020-07-31 15:47:00,306-INFO: iter: 1740, lr: 0.000054, 'loss': '100.238541', time: 0.290, eta: 3:53:16 2020-07-31 15:47:06,376-INFO: iter: 1760, lr: 0.000055, 'loss': '97.934555', time: 0.309, eta: 4:08:28 2020-07-31 15:47:12,769-INFO: iter: 1780, lr: 0.000056, 'loss': '95.060623', time: 0.326, eta: 4:21:45 2020-07-31 15:47:19,247-INFO: iter: 1800, lr: 0.000056, 'loss': '113.702446', time: 0.314, eta: 4:11:55 2020-07-31 15:47:25,983-INFO: iter: 1820, lr: 0.000057, 'loss': '94.452484', time: 0.337, eta: 4:30:55 2020-07-31 15:47:32,322-INFO: iter: 1840, lr: 0.000058, 'loss': '97.577782', time: 0.308, eta: 4:06:59 2020-07-31 15:47:40,725-INFO: iter: 1860, lr: 0.000058, 'loss': '108.748138', time: 0.431, eta: 5:45:42 ^C 2020-07-31 15:47:41,125-INFO: KeyboardInterrupt: main proc 206 exit, kill subprocess [295, 298, 296, 302, 300, 299, 297, 301]
查看loss和精度值曲线
启动训练后，可以使用Visualdl可以在训练期间实时观察loss和精度值曲线。
查看方法：新打开一个新的网页窗口，在地址栏参考以下格式输入网址。
格式：AI studio链接+visualdl，比如：https://aistudio.baidu.com/bdvgpu/user/166411/627226/visualdl
一键式模型推理
当我们训练完成后，在项目文件的output中可以看到我们生成的模型文件，在代码中我们设置模型每迭代200步保存一次，训练总共有50000步，因此会生成250个训练文件，由于我们使用 --eval参数进行边训练边测试，因此我们可以获得训练过程中最好的模型文件。
PaddleDetection给出的模型预测脚本是infer.py，可通过此脚本，使用训练好的模型对指定图像进行推理预测。该脚本在项目文件夹tools里面。PaddleDetection提供了两种预测方式，第一种是单张图片预测，第二种是以一个文件夹中的图片进行预测。使用–infer_img=demo/xxx.jpg为单张图片进行预测，使用–infer_dir=demo为用一个文件夹进行预测。其中demo是放置测试图片的地方。
infer.py默认加载上面一键式训练的最终权重output/yolov3_darknet/model_final, 可通过-o weights=指定模型权重路径，若指定路径为url，将自动下载解压并加载。本示例中自动下载并加载PaddleDetection发布的权重。
预测结果默认存储于output目录，可通过如下方式可视化预测结果。
# ! python tools/infer.py -c ../../work/yolov3_darknet.yml --infer_img=../../work/dog.jpg
! python tools/infer.py -c ../../work/yolov3_darknet.yml --infer_img=../../work/dog.jpg -o weights=https://paddlemodels.bj.bcebos.com/object_detection/yolov3_darknet.tarW0731 15:48:09.248399 652 device_context.cc:252] Please NOTE: device: 0, CUDA Capability: 70, Driver API Version: 10.1, Runtime API Version: 9.0 W0731 15:48:09.252437 652 device_context.cc:260] device: 0, cuDNN Version: 7.3. 2020-07-31 15:48:11,873-WARNING: /home/aistudio/.cache/paddle/weights/yolov3_darknet.pdparams not found, try to load model file saved with [ save_params, save_persistables, save_vars ] 2020-07-31 15:48:12,584-INFO: Not found annotation file annotations/instances_val2017.json, load coco17 categories. 2020-07-31 15:48:12,800-INFO: Infer iter 0 2020-07-31 15:48:12,817-INFO: Detection bbox results save in output/dog.jpg
%matplotlib inline
import matplotlib.pyplot as plt
import cv2infer_img = cv2.imread("output/dog.jpg")
plt.figure(figsize=(15,10))
plt.imshow(cv2.cvtColor(infer_img, cv2.COLOR_BGR2RGB))
plt.show()
<Figure size 1080x720 with 1 Axes>
一键式模型部署
模型导出
模型部署须将模型导出为Paddle预测库支持的格式。PaddleDetection提供了模型导出脚本tools/export_model.py，通过此脚本可以一键式下载PaddleDetection发布的yolov3_darknet模型权重并完成模型导出。导出的模型存储于output/yolov3_darknet目录。
! python tools/export_model.py -c configs/yolov3_darknet.yml -o weights=https://paddlemodels.bj.bcebos.com/object_detection/yolov3_darknet.tar2020-07-31 15:52:56,613-WARNING: /home/aistudio/.cache/paddle/weights/yolov3_darknet.pdparams not found, try to load model file saved with [ save_params, save_persistables, save_vars ] 2020-07-31 15:52:57,611-INFO: save_inference_model pruned unused feed variables im_id 2020-07-31 15:52:57,611-INFO: Export inference model to output/yolov3_darknet, input: ['image', 'im_size'], output: ['multiclass_nms_0.tmp_0']... 2020-07-31 15:52:58,548-INFO: Not found annotation file annotations/instances_val2017.json, load coco17 categories. 2020-07-31 15:52:58,552-INFO: Export inference config file to output/yolov3_darknet/infer_cfg.yml
图像推理
PaddleDetection在deploy/python文件夹下提供模型推理的脚本infer.py,可以实现Python端调用Paddle预测库完成图像和视频的推理，并可视化推理结果。
! python deploy/python/infer.py --model_dir=output/yolov3_darknet/ --image_file=../../work/dog.jpg --use_gpu=True----------- Running Arguments ----------- image_file: ../../work/dog.jpg model_dir: output/yolov3_darknet/ output_dir: output run_benchmark: False run_mode: fluid threshold: 0.5 use_gpu: True video_file: ------------------------------------------ ----------- Model Configuration ----------- Model Arch: YOLO Use Padddle Executor: False Transform Order: --transform op: Resize --transform op: Normalize --transform op: Permute -------------------------------------------- Inference: 22.216081619262695 ms per batch image class_id:1, confidence:0.96,left_top:[122.40,125.05], right_bottom:[560.99,435.19] class_id:7, confidence:0.85,left_top:[471.86,80.75], right_bottom:[685.44,167.01] class_id:16, confidence:0.98,left_top:[131.98,224.07], right_bottom:[316.92,545.75] save result to: output/dog.jpg
%matplotlib inline
import matplotlib.pyplot as plt
import cv2infer_img = cv2.imread("output/dog.jpg")
plt.figure(figsize=(15,10))
plt.imshow(cv2.cvtColor(infer_img, cv2.COLOR_BGR2RGB))
plt.show()
<Figure size 1080x720 with 1 Axes>
视频推理
同样使用deploy/python/infer.py进行视频推理。视频推理结果不能在AIStudio上展示，可以在data/PaddleDetection/output目录下载预测结果视频到本地播放，确认预测效果。
为了在视频流预测中效果更好，可以设置
--thresh=0.2，这样score >= 0.2的预测框均画在可视化结果中。
! python deploy/python/infer.py --model_dir=output/yolov3_darknet/ --video_file=../../work/test.mp4 --use_gpu=True --thresh=0.2Inference: 20.867347717285156 ms per batch image class_id:0, confidence:0.92,left_top:[802.62,152.84], right_bottom:[863.96,284.90] class_id:17, confidence:0.99,left_top:[706.12,171.99], right_bottom:[938.53,364.24] class_id:17, confidence:0.97,left_top:[140.06,251.72], right_bottom:[356.46,416.49] class_id:17, confidence:0.95,left_top:[432.89,222.03], right_bottom:[629.28,376.07] class_id:17, confidence:0.94,left_top:[549.35,230.51], right_bottom:[709.38,367.12] class_id:17, confidence:0.75,left_top:[0.00,257.36], right_bottom:[86.34,417.62] detect frame:145 Inference: 20.87688446044922 ms per batch image class_id:0, confidence:0.75,left_top:[816.19,139.67], right_bottom:[874.47,282.75] class_id:17, confidence:0.99,left_top:[710.34,161.08], right_bottom:[949.00,355.40] class_id:17, confidence:0.95,left_top:[457.67,219.18], right_bottom:[637.15,367.83] class_id:17, confidence:0.93,left_top:[140.75,248.25], right_bottom:[363.70,412.64] class_id:17, confidence:0.85,left_top:[584.36,220.84], right_bottom:[717.28,359.18] class_id:17, confidence:0.75,left_top:[0.00,258.11], right_bottom:[82.99,413.29] detect frame:146 Inference: 20.86806297302246 ms per batch image class_id:0, confidence:0.87,left_top:[838.77,122.24], right_bottom:[897.83,271.42] class_id:17, confidence:0.99,left_top:[739.84,148.93], right_bottom:[971.95,349.16] class_id:17, confidence:0.96,left_top:[583.26,214.90], right_bottom:[743.49,359.32] class_id:17, confidence:0.95,left_top:[476.78,215.65], right_bottom:[654.16,365.25] class_id:17, confidence:0.95,left_top:[0.00,252.64], right_bottom:[98.08,410.19] class_id:17, confidence:0.89,left_top:[177.31,242.36], right_bottom:[382.36,402.14] detect frame:147 Inference: 20.834684371948242 ms per batch image class_id:0, confidence:0.88,left_top:[855.00,117.39], right_bottom:[912.18,269.93] class_id:17, confidence:0.99,left_top:[744.04,143.35], right_bottom:[989.51,347.78] class_id:17, confidence:0.96,left_top:[488.78,205.83], right_bottom:[671.08,357.99] class_id:17, confidence:0.93,left_top:[1.38,242.37], right_bottom:[107.43,394.92] class_id:17, confidence:0.90,left_top:[606.82,202.20], right_bottom:[759.10,346.42] class_id:17, confidence:0.74,left_top:[173.84,229.31], right_bottom:[405.05,395.68] detect frame:148 Inference: 20.95198631286621 ms per batch image class_id:0, confidence:0.82,left_top:[866.98,117.93], right_bottom:[930.13,273.13] class_id:17, confidence:0.99,left_top:[761.65,144.62], right_bottom:[1010.87,353.06] class_id:17, confidence:0.97,left_top:[499.06,205.60], right_bottom:[689.92,357.20] class_id:17, confidence:0.96,left_top:[205.44,225.30], right_bottom:[413.25,390.66] class_id:17, confidence:0.93,left_top:[612.58,201.05], right_bottom:[777.56,346.86] class_id:17, confidence:0.92,left_top:[0.00,237.70], right_bottom:[117.70,393.22] detect frame:149 Inference: 20.865678787231445 ms per batch image class_id:0, confidence:0.92,left_top:[887.07,118.34], right_bottom:[942.18,274.42] class_id:17, confidence:0.99,left_top:[775.14,144.32], right_bottom:[1030.93,360.31] class_id:17, confidence:0.98,left_top:[218.87,216.38], right_bottom:[429.78,383.15] class_id:17, confidence:0.93,left_top:[0.00,237.04], right_bottom:[129.51,388.95] class_id:17, confidence:0.90,left_top:[505.64,204.31], right_bottom:[699.70,357.55] class_id:17, confidence:0.84,left_top:[639.16,202.27], right_bottom:[792.76,347.65] detect frame:150 Inference: 20.720243453979492 ms per batch image class_id:0, confidence:0.78,left_top:[901.95,120.48], right_bottom:[964.01,279.41] class_id:17, confidence:0.99,left_top:[787.41,145.08], right_bottom:[1045.18,361.12] class_id:17, confidence:0.99,left_top:[234.14,212.56], right_bottom:[445.91,380.77] class_id:17, confidence:0.95,left_top:[524.80,205.21], right_bottom:[729.90,361.79] class_id:17, confidence:0.92,left_top:[0.81,224.22], right_bottom:[142.62,376.47] class_id:17, confidence:0.87,left_top:[636.71,203.00], right_bottom:[809.56,348.67] detect frame:151 Inference: 20.743608474731445 ms per batch image class_id:0, confidence:0.73,left_top:[923.29,128.46], right_bottom:[976.99,293.44] class_id:17, confidence:0.98,left_top:[0.00,234.39], right_bottom:[152.19,397.33] class_id:17, confidence:0.98,left_top:[814.06,155.82], right_bottom:[1051.07,365.46] class_id:17, confidence:0.96,left_top:[244.49,215.97], right_bottom:[452.61,379.61] class_id:17, confidence:0.89,left_top:[540.92,213.95], right_bottom:[737.92,364.77] class_id:17, confidence:0.72,left_top:[669.61,218.10], right_bottom:[828.76,359.49] detect frame:152 Inference: 20.807266235351562 ms per batch image class_id:0, confidence:0.80,left_top:[933.73,118.50], right_bottom:[986.56,276.68] class_id:17, confidence:0.99,left_top:[827.28,143.34], right_bottom:[1073.93,365.56] class_id:17, confidence:0.97,left_top:[558.65,203.50], right_bottom:[744.98,356.70] class_id:17, confidence:0.97,left_top:[0.00,231.73], right_bottom:[161.34,389.45] class_id:17, confidence:0.96,left_top:[251.65,211.89], right_bottom:[463.97,371.45] class_id:17, confidence:0.94,left_top:[666.61,203.63], right_bottom:[836.48,357.46] detect frame:153 Inference: 20.956993103027344 ms per batch image class_id:0, confidence:0.79,left_top:[935.56,131.87], right_bottom:[997.06,292.52] class_id:17, confidence:0.99,left_top:[822.86,167.34], right_bottom:[1074.22,370.23] class_id:17, confidence:0.98,left_top:[259.78,215.11], right_bottom:[474.61,374.99] class_id:17, confidence:0.98,left_top:[0.00,231.90], right_bottom:[163.87,389.45] class_id:17, confidence:0.93,left_top:[569.98,216.16], right_bottom:[748.93,364.95] class_id:17, confidence:0.93,left_top:[673.06,217.12], right_bottom:[845.91,364.70] detect frame:154 Inference: 20.941734313964844 ms per batch image class_id:0, confidence:0.79,left_top:[941.19,148.42], right_bottom:[996.16,295.95] class_id:17, confidence:0.99,left_top:[828.15,178.69], right_bottom:[1076.98,377.58] class_id:17, confidence:0.98,left_top:[268.58,229.89], right_bottom:[476.39,386.61] class_id:17, confidence:0.97,left_top:[0.00,242.15], right_bottom:[166.64,396.62] class_id:17, confidence:0.94,left_top:[572.39,237.29], right_bottom:[752.99,374.16] class_id:17, confidence:0.90,left_top:[683.07,226.55], right_bottom:[848.38,370.89] detect frame:155 Inference: 20.961999893188477 ms per batch image class_id:0, confidence:0.84,left_top:[939.24,139.33], right_bottom:[1001.48,295.57] class_id:17, confidence:0.99,left_top:[831.11,172.95], right_bottom:[1078.63,375.02] class_id:17, confidence:0.98,left_top:[276.89,223.74], right_bottom:[480.77,389.19] class_id:17, confidence:0.98,left_top:[1.84,245.51], right_bottom:[167.54,403.44] class_id:17, confidence:0.95,left_top:[567.56,226.48], right_bottom:[756.09,375.07] class_id:17, confidence:0.93,left_top:[685.48,217.36], right_bottom:[849.85,371.53] detect frame:156 Inference: 20.688295364379883 ms per batch image class_id:0, confidence:0.65,left_top:[938.74,153.05], right_bottom:[1004.66,308.02] class_id:17, confidence:0.99,left_top:[838.10,180.77], right_bottom:[1084.14,381.98] class_id:17, confidence:0.99,left_top:[286.75,244.39], right_bottom:[489.68,413.23] class_id:17, confidence:0.99,left_top:[4.48,279.29], right_bottom:[179.40,430.17] class_id:17, confidence:0.96,left_top:[570.94,239.95], right_bottom:[762.22,389.61] class_id:17, confidence:0.85,left_top:[680.30,231.70], right_bottom:[855.64,385.25] detect frame:157 Inference: 20.906448364257812 ms per batch image class_id:0, confidence:0.74,left_top:[938.64,153.78], right_bottom:[1005.18,307.02] class_id:17, confidence:0.99,left_top:[834.47,181.22], right_bottom:[1088.15,386.77] class_id:17, confidence:0.99,left_top:[289.24,247.86], right_bottom:[493.85,419.90] class_id:17, confidence:0.97,left_top:[3.56,286.88], right_bottom:[181.05,441.72] class_id:17, confidence:0.88,left_top:[571.10,240.67], right_bottom:[764.11,394.44] class_id:17, confidence:0.71,left_top:[677.54,230.66], right_bottom:[855.22,391.93] detect frame:158 Inference: 20.777225494384766 ms per batch image class_id:0, confidence:0.92,left_top:[929.25,142.19], right_bottom:[997.37,301.30] class_id:17, confidence:0.99,left_top:[825.87,175.61], right_bottom:[1089.03,393.02] class_id:17, confidence:0.99,left_top:[284.23,245.05], right_bottom:[484.42,424.60] class_id:17, confidence:0.98,left_top:[4.54,287.07], right_bottom:[170.83,455.74] class_id:17, confidence:0.92,left_top:[570.58,237.92], right_bottom:[761.40,398.05] class_id:17, confidence:0.82,left_top:[677.82,227.75], right_bottom:[852.57,393.83] detect frame:159 Inference: 20.716428756713867 ms per batch image class_id:0, confidence:0.95,left_top:[923.66,157.68], right_bottom:[991.56,311.10] class_id:17, confidence:0.99,left_top:[832.73,183.93], right_bottom:[1081.99,397.45] class_id:17, confidence:0.99,left_top:[0.78,306.62], right_bottom:[160.36,471.07] class_id:17, confidence:0.98,left_top:[280.63,261.84], right_bottom:[478.77,436.38] class_id:17, confidence:0.86,left_top:[561.40,244.43], right_bottom:[764.13,407.03] class_id:17, confidence:0.78,left_top:[679.85,235.72], right_bottom:[852.38,400.19] detect frame:160 Inference: 20.795106887817383 ms per batch image class_id:0, confidence:0.89,left_top:[921.44,163.77], right_bottom:[990.80,317.62] class_id:17, confidence:0.99,left_top:[0.00,312.73], right_bottom:[162.41,468.77] class_id:17, confidence:0.99,left_top:[822.68,183.35], right_bottom:[1080.17,412.09] class_id:17, confidence:0.97,left_top:[276.95,263.92], right_bottom:[474.51,440.27] class_id:17, confidence:0.94,left_top:[572.71,251.71], right_bottom:[755.38,406.70] class_id:17, confidence:0.87,left_top:[676.37,241.57], right_bottom:[848.72,413.66] detect frame:161 Inference: 20.885467529296875 ms per batch image class_id:0, confidence:0.88,left_top:[918.45,190.08], right_bottom:[985.08,347.73] class_id:17, confidence:0.98,left_top:[817.18,210.49], right_bottom:[1081.67,418.23] class_id:17, confidence:0.98,left_top:[571.71,269.29], right_bottom:[749.93,422.49] class_id:17, confidence:0.97,left_top:[0.00,329.73], right_bottom:[152.10,476.67] class_id:17, confidence:0.97,left_top:[268.58,286.23], right_bottom:[469.80,452.94] class_id:17, confidence:0.95,left_top:[678.98,262.81], right_bottom:[845.11,425.57] detect frame:162 Inference: 20.87545394897461 ms per batch image class_id:0, confidence:0.81,left_top:[916.12,191.69], right_bottom:[982.23,345.36] class_id:17, confidence:0.99,left_top:[799.00,212.82], right_bottom:[1064.20,419.77] class_id:17, confidence:0.98,left_top:[565.26,269.67], right_bottom:[740.69,430.49] class_id:17, confidence:0.98,left_top:[0.00,331.68], right_bottom:[146.33,486.03] class_id:17, confidence:0.95,left_top:[672.26,262.87], right_bottom:[837.12,429.38] class_id:17, confidence:0.95,left_top:[268.72,285.04], right_bottom:[469.77,461.68] detect frame:163 Inference: 20.832300186157227 ms per batch image class_id:0, confidence:0.92,left_top:[918.52,201.04], right_bottom:[984.30,354.17] class_id:17, confidence:0.99,left_top:[813.66,221.80], right_bottom:[1056.22,426.19] class_id:17, confidence:0.98,left_top:[561.14,275.75], right_bottom:[745.68,434.68] class_id:17, confidence:0.98,left_top:[0.00,334.99], right_bottom:[146.95,490.60] class_id:17, confidence:0.98,left_top:[267.06,297.27], right_bottom:[473.87,468.82] class_id:17, confidence:0.96,left_top:[669.45,266.68], right_bottom:[842.07,437.16] detect frame:164 Inference: 20.642757415771484 ms per batch image class_id:0, confidence:0.93,left_top:[920.82,214.68], right_bottom:[976.48,358.69] class_id:17, confidence:0.99,left_top:[811.20,239.20], right_bottom:[1056.47,430.97] class_id:17, confidence:0.98,left_top:[256.56,304.44], right_bottom:[466.01,469.46] class_id:17, confidence:0.97,left_top:[558.08,290.63], right_bottom:[735.51,442.88] class_id:17, confidence:0.94,left_top:[0.00,345.98], right_bottom:[130.30,493.75] class_id:17, confidence:0.92,left_top:[670.98,283.09], right_bottom:[833.85,443.27] detect frame:165 Inference: 20.650386810302734 ms per batch image class_id:0, confidence:0.84,left_top:[917.78,197.79], right_bottom:[982.11,354.24] class_id:17, confidence:0.99,left_top:[801.93,222.88], right_bottom:[1056.67,428.93] class_id:17, confidence:0.99,left_top:[258.03,298.32], right_bottom:[473.41,470.37] class_id:17, confidence:0.96,left_top:[556.74,277.08], right_bottom:[739.97,437.30] class_id:17, confidence:0.94,left_top:[0.00,340.20], right_bottom:[130.22,499.83] class_id:17, confidence:0.86,left_top:[667.54,264.97], right_bottom:[837.91,439.04] detect frame:166 Inference: 20.69234848022461 ms per batch image class_id:0, confidence:0.64,left_top:[916.95,218.44], right_bottom:[976.39,362.24] class_id:17, confidence:0.99,left_top:[799.36,251.46], right_bottom:[1054.29,443.97] class_id:17, confidence:0.99,left_top:[242.93,315.58], right_bottom:[467.14,479.19] class_id:17, confidence:0.96,left_top:[542.97,295.47], right_bottom:[739.68,448.14] class_id:17, confidence:0.89,left_top:[0.00,348.70], right_bottom:[119.57,489.78] class_id:17, confidence:0.79,left_top:[674.74,290.87], right_bottom:[832.37,443.79] detect frame:167 Inference: 20.741701126098633 ms per batch image class_id:0, confidence:0.85,left_top:[917.23,197.67], right_bottom:[982.32,356.28] class_id:17, confidence:0.99,left_top:[799.38,227.39], right_bottom:[1060.95,443.85] class_id:17, confidence:0.98,left_top:[251.60,298.06], right_bottom:[465.37,474.57] class_id:17, confidence:0.95,left_top:[0.00,336.42], right_bottom:[119.22,486.33] class_id:17, confidence:0.94,left_top:[538.62,284.12], right_bottom:[736.85,444.82] class_id:17, confidence:0.74,left_top:[670.19,277.53], right_bottom:[832.05,452.04] detect frame:168 Inference: 20.80535888671875 ms per batch image class_id:0, confidence:0.61,left_top:[916.06,204.56], right_bottom:[980.56,368.86] class_id:17, confidence:0.99,left_top:[797.81,241.30], right_bottom:[1056.59,447.93] class_id:17, confidence:0.97,left_top:[247.68,302.53], right_bottom:[462.88,474.22] class_id:17, confidence:0.95,left_top:[539.62,289.55], right_bottom:[739.79,450.49] class_id:17, confidence:0.95,left_top:[0.00,334.61], right_bottom:[112.79,491.45] class_id:17, confidence:0.65,left_top:[620.14,286.36], right_bottom:[814.81,454.00] detect frame:169 Inference: 20.83730697631836 ms per batch image class_id:0, confidence:0.82,left_top:[915.70,207.41], right_bottom:[978.52,371.44] class_id:17, confidence:0.99,left_top:[792.57,243.16], right_bottom:[1053.10,452.50] class_id:17, confidence:0.96,left_top:[0.00,335.05], right_bottom:[112.73,490.30] class_id:17, confidence:0.96,left_top:[537.30,289.80], right_bottom:[738.29,452.77] class_id:17, confidence:0.96,left_top:[243.24,303.30], right_bottom:[454.95,477.60] class_id:17, confidence:0.82,left_top:[642.85,285.41], right_bottom:[820.20,452.87] detect frame:170 Inference: 20.904541015625 ms per batch image class_id:0, confidence:0.75,left_top:[907.36,213.07], right_bottom:[970.97,373.98] class_id:17, confidence:0.98,left_top:[790.33,243.34], right_bottom:[1041.94,454.11] class_id:17, confidence:0.95,left_top:[0.00,334.68], right_bottom:[110.09,492.09] class_id:17, confidence:0.95,left_top:[238.27,309.78], right_bottom:[451.56,480.98] class_id:17, confidence:0.93,left_top:[536.94,297.02], right_bottom:[727.27,454.22] class_id:17, confidence:0.92,left_top:[642.29,289.70], right_bottom:[813.71,460.91] detect frame:171 Inference: 20.809650421142578 ms per batch image class_id:0, confidence:0.92,left_top:[894.92,222.04], right_bottom:[950.42,371.81] class_id:17, confidence:0.99,left_top:[788.00,249.72], right_bottom:[1033.72,452.96] class_id:17, confidence:0.98,left_top:[219.25,322.11], right_bottom:[441.22,497.01] class_id:17, confidence:0.93,left_top:[633.30,299.88], right_bottom:[796.53,464.58] class_id:17, confidence:0.89,left_top:[0.00,359.94], right_bottom:[100.89,517.94] class_id:17, confidence:0.87,left_top:[530.24,310.65], right_bottom:[707.98,463.50] detect frame:172 Inference: 20.681381225585938 ms per batch image class_id:0, confidence:0.91,left_top:[891.25,209.89], right_bottom:[947.06,366.77] class_id:17, confidence:0.99,left_top:[784.12,238.19], right_bottom:[1025.84,451.10] class_id:17, confidence:0.94,left_top:[0.00,362.79], right_bottom:[102.39,525.95] class_id:17, confidence:0.93,left_top:[637.90,286.17], right_bottom:[791.92,462.42] class_id:17, confidence:0.89,left_top:[524.70,306.23], right_bottom:[699.93,462.16] class_id:17, confidence:0.88,left_top:[222.05,319.58], right_bottom:[438.69,499.62] detect frame:173 Inference: 20.882129669189453 ms per batch image class_id:0, confidence:0.87,left_top:[893.41,202.19], right_bottom:[948.06,358.76] class_id:17, confidence:0.99,left_top:[778.31,235.25], right_bottom:[1024.53,454.18] class_id:17, confidence:0.90,left_top:[0.00,351.93], right_bottom:[99.39,513.87] class_id:17, confidence:0.87,left_top:[615.35,274.04], right_bottom:[789.74,465.15] class_id:17, confidence:0.86,left_top:[219.89,312.92], right_bottom:[436.83,497.55] class_id:17, confidence:0.77,left_top:[511.70,313.69], right_bottom:[682.60,464.20] detect frame:174 Inference: 20.795345306396484 ms per batch image class_id:0, confidence:0.95,left_top:[890.05,222.36], right_bottom:[943.93,370.66] class_id:17, confidence:0.99,left_top:[768.73,251.97], right_bottom:[1026.28,452.04] class_id:17, confidence:0.98,left_top:[219.94,320.58], right_bottom:[438.63,496.29] class_id:17, confidence:0.90,left_top:[512.65,307.64], right_bottom:[721.69,466.35] class_id:17, confidence:0.87,left_top:[0.00,355.59], right_bottom:[95.20,509.35] class_id:17, confidence:0.75,left_top:[613.96,292.66], right_bottom:[786.39,454.43] detect frame:175 Inference: 20.80988883972168 ms per batch image class_id:0, confidence:0.88,left_top:[873.97,219.98], right_bottom:[938.57,371.83] class_id:17, confidence:1.00,left_top:[754.99,250.13], right_bottom:[1020.82,448.67] class_id:17, confidence:0.97,left_top:[220.32,317.15], right_bottom:[432.99,494.85] class_id:17, confidence:0.87,left_top:[0.00,349.77], right_bottom:[89.27,507.61] class_id:17, confidence:0.83,left_top:[494.88,315.40], right_bottom:[680.15,464.76] class_id:17, confidence:0.76,left_top:[601.65,289.55], right_bottom:[774.12,455.94] detect frame:176 Inference: 20.653724670410156 ms per batch image class_id:0, confidence:0.86,left_top:[862.00,211.04], right_bottom:[930.49,363.91] class_id:17, confidence:1.00,left_top:[752.94,234.85], right_bottom:[1020.67,442.18] class_id:17, confidence:0.96,left_top:[212.85,320.81], right_bottom:[416.05,496.45] class_id:17, confidence:0.95,left_top:[0.00,365.46], right_bottom:[77.02,512.72] class_id:17, confidence:0.88,left_top:[492.53,311.05], right_bottom:[671.09,466.85] class_id:17, confidence:0.74,left_top:[580.55,290.81], right_bottom:[752.69,459.88] detect frame:177 Inference: 20.68305015563965 ms per batch image class_id:0, confidence:0.92,left_top:[869.31,216.49], right_bottom:[935.01,368.10] class_id:17, confidence:0.99,left_top:[751.74,246.82], right_bottom:[1027.97,446.06] class_id:17, confidence:0.96,left_top:[215.39,319.92], right_bottom:[416.33,491.10] class_id:17, confidence:0.96,left_top:[582.33,297.49], right_bottom:[747.23,476.42] class_id:17, confidence:0.92,left_top:[0.00,356.82], right_bottom:[79.64,500.63] class_id:17, confidence:0.73,left_top:[497.86,313.46], right_bottom:[670.79,469.24] detect frame:178 Inference: 20.751953125 ms per batch image class_id:0, confidence:0.80,left_top:[866.25,221.63], right_bottom:[932.30,371.06] class_id:17, confidence:0.99,left_top:[751.70,250.46], right_bottom:[1025.53,453.56] class_id:17, confidence:0.97,left_top:[210.20,317.92], right_bottom:[413.32,490.57] class_id:17, confidence:0.93,left_top:[581.25,300.32], right_bottom:[744.38,474.15] class_id:17, confidence:0.90,left_top:[0.00,351.47], right_bottom:[75.48,497.19] class_id:17, confidence:0.69,left_top:[499.75,313.34], right_bottom:[668.84,468.92] detect frame:179 Inference: 20.84946632385254 ms per batch image class_id:0, confidence:0.92,left_top:[861.89,235.50], right_bottom:[920.65,383.67] class_id:17, confidence:0.99,left_top:[754.73,257.82], right_bottom:[1016.98,455.23] class_id:17, confidence:0.98,left_top:[194.40,323.02], right_bottom:[402.67,490.85] class_id:17, confidence:0.91,left_top:[491.83,320.41], right_bottom:[668.90,470.89] class_id:17, confidence:0.90,left_top:[582.15,306.09], right_bottom:[739.29,474.74] class_id:17, confidence:0.73,left_top:[1.99,353.75], right_bottom:[70.61,497.25] detect frame:180 Inference: 20.71404457092285 ms per batch image class_id:0, confidence:0.87,left_top:[863.44,221.61], right_bottom:[926.80,379.62] class_id:17, confidence:0.99,left_top:[743.07,249.23], right_bottom:[1024.90,454.77] class_id:17, confidence:0.97,left_top:[196.92,312.00], right_bottom:[408.95,487.22] class_id:17, confidence:0.84,left_top:[493.00,312.42], right_bottom:[670.55,472.48] class_id:17, confidence:0.78,left_top:[576.35,296.71], right_bottom:[741.67,473.95] class_id:17, confidence:0.65,left_top:[2.07,335.69], right_bottom:[71.00,482.60] detect frame:181 Inference: 20.897388458251953 ms per batch image class_id:0, confidence:0.85,left_top:[861.79,244.88], right_bottom:[915.93,387.79] class_id:17, confidence:0.99,left_top:[753.61,265.18], right_bottom:[1009.66,461.92] class_id:17, confidence:0.97,left_top:[196.04,317.42], right_bottom:[401.26,483.02] class_id:17, confidence:0.85,left_top:[574.29,303.64], right_bottom:[733.32,477.01] class_id:17, confidence:0.63,left_top:[0.00,330.19], right_bottom:[64.25,485.48] class_id:17, confidence:0.63,left_top:[482.03,315.53], right_bottom:[685.51,475.10] detect frame:182 Inference: 20.75362205505371 ms per batch image class_id:0, confidence:0.82,left_top:[859.62,235.72], right_bottom:[911.69,389.75] class_id:17, confidence:0.98,left_top:[741.04,261.33], right_bottom:[996.92,447.92] class_id:17, confidence:0.93,left_top:[192.29,313.07], right_bottom:[393.00,481.47] class_id:17, confidence:0.80,left_top:[483.02,306.16], right_bottom:[692.59,469.97] class_id:17, confidence:0.74,left_top:[580.21,300.48], right_bottom:[727.95,471.90] class_id:17, confidence:0.36,left_top:[0.00,328.56], right_bottom:[62.82,489.02] detect frame:183 Inference: 20.813465118408203 ms per batch image class_id:0, confidence:0.87,left_top:[858.16,232.28], right_bottom:[912.60,387.36] class_id:17, confidence:0.99,left_top:[739.13,259.35], right_bottom:[997.19,450.34] class_id:17, confidence:0.95,left_top:[186.79,315.03], right_bottom:[392.92,487.83] class_id:17, confidence:0.80,left_top:[479.38,304.78], right_bottom:[684.31,469.36] class_id:17, confidence:0.74,left_top:[580.84,294.52], right_bottom:[728.40,452.77] class_id:17, confidence:0.41,left_top:[0.00,340.89], right_bottom:[60.76,500.07] detect frame:184 Inference: 20.818471908569336 ms per batch image class_id:0, confidence:0.88,left_top:[859.67,236.91], right_bottom:[913.60,385.23] class_id:17, confidence:0.99,left_top:[734.56,262.95], right_bottom:[997.71,449.95] class_id:17, confidence:0.94,left_top:[177.98,317.82], right_bottom:[393.25,485.45] class_id:17, confidence:0.89,left_top:[479.36,305.05], right_bottom:[698.79,465.92] class_id:17, confidence:0.58,left_top:[580.08,294.64], right_bottom:[723.63,457.47] class_id:17, confidence:0.43,left_top:[1.32,339.55], right_bottom:[50.87,508.93] detect frame:185 Inference: 20.677566528320312 ms per batch image class_id:0, confidence:0.85,left_top:[857.05,229.71], right_bottom:[911.99,385.11] class_id:17, confidence:0.99,left_top:[733.44,264.24], right_bottom:[994.63,448.74] class_id:17, confidence:0.91,left_top:[175.81,309.71], right_bottom:[387.42,481.19] class_id:17, confidence:0.72,left_top:[576.17,292.34], right_bottom:[722.31,458.71] class_id:17, confidence:0.62,left_top:[471.35,308.98], right_bottom:[641.92,467.92] class_id:17, confidence:0.27,left_top:[2.01,331.85], right_bottom:[46.29,490.13] detect frame:186 Inference: 20.761966705322266 ms per batch image class_id:0, confidence:0.87,left_top:[858.66,219.14], right_bottom:[912.83,374.43] class_id:17, confidence:0.99,left_top:[729.17,256.26], right_bottom:[1000.28,446.00] class_id:17, confidence:0.88,left_top:[479.27,301.76], right_bottom:[646.19,469.96] class_id:17, confidence:0.85,left_top:[583.24,290.77], right_bottom:[717.95,455.70] class_id:17, confidence:0.78,left_top:[232.18,301.52], right_bottom:[388.17,480.70] class_id:17, confidence:0.39,left_top:[0.54,333.95], right_bottom:[42.13,475.75] detect frame:187 Inference: 20.71404457092285 ms per batch image class_id:0, confidence:0.91,left_top:[859.49,219.11], right_bottom:[913.58,375.40] class_id:17, confidence:0.99,left_top:[737.27,252.63], right_bottom:[997.34,444.99] class_id:17, confidence:0.85,left_top:[485.32,299.31], right_bottom:[649.04,466.46] class_id:17, confidence:0.84,left_top:[232.22,303.01], right_bottom:[386.78,482.73] class_id:17, confidence:0.75,left_top:[585.78,294.98], right_bottom:[716.66,447.14] class_id:17, confidence:0.56,left_top:[0.89,337.86], right_bottom:[40.57,472.54] class_id:17, confidence:0.21,left_top:[141.90,333.32], right_bottom:[293.00,480.73] detect frame:188 Inference: 20.679473876953125 ms per batch image class_id:0, confidence:0.89,left_top:[856.26,219.73], right_bottom:[918.40,377.85] class_id:17, confidence:0.99,left_top:[736.17,250.61], right_bottom:[998.16,445.21] class_id:17, confidence:0.69,left_top:[145.23,309.88], right_bottom:[365.18,489.51] class_id:17, confidence:0.69,left_top:[495.40,299.00], right_bottom:[663.09,466.07] class_id:17, confidence:0.56,left_top:[0.00,337.83], right_bottom:[38.47,473.57] class_id:17, confidence:0.52,left_top:[588.01,296.28], right_bottom:[715.08,442.31] detect frame:189 Inference: 21.020889282226562 ms per batch image class_id:0, confidence:0.91,left_top:[858.65,215.29], right_bottom:[913.63,373.44] class_id:17, confidence:0.99,left_top:[732.64,246.24], right_bottom:[998.35,442.55] class_id:17, confidence:0.92,left_top:[475.78,290.46], right_bottom:[646.54,460.44] class_id:17, confidence:0.82,left_top:[584.77,295.17], right_bottom:[711.01,447.86] class_id:17, confidence:0.70,left_top:[141.57,294.58], right_bottom:[368.76,489.52] class_id:17, confidence:0.31,left_top:[1.89,332.65], right_bottom:[40.09,481.07] detect frame:190 Inference: 20.846128463745117 ms per batch image class_id:0, confidence:0.77,left_top:[854.90,204.91], right_bottom:[916.70,369.91] class_id:17, confidence:0.99,left_top:[734.30,243.48], right_bottom:[1001.12,444.60] class_id:17, confidence:0.80,left_top:[469.45,281.88], right_bottom:[647.73,457.04] class_id:17, confidence:0.75,left_top:[544.80,285.54], right_bottom:[714.69,453.03] class_id:17, confidence:0.69,left_top:[142.36,288.87], right_bottom:[369.79,481.18] class_id:17, confidence:0.37,left_top:[1.09,325.67], right_bottom:[43.06,488.42] class_id:17, confidence:0.22,left_top:[145.08,344.11], right_bottom:[288.63,476.50] detect frame:191 Inference: 20.681142807006836 ms per batch image class_id:0, confidence:0.94,left_top:[863.36,214.75], right_bottom:[916.22,373.43] class_id:17, confidence:0.99,left_top:[735.40,250.39], right_bottom:[1002.28,440.74] class_id:17, confidence:0.85,left_top:[152.36,289.40], right_bottom:[375.03,478.12] class_id:17, confidence:0.76,left_top:[540.65,290.95], right_bottom:[720.96,451.81] class_id:17, confidence:0.71,left_top:[465.11,293.00], right_bottom:[650.05,455.52] class_id:17, confidence:0.48,left_top:[0.00,326.64], right_bottom:[43.10,481.49] detect frame:192 Inference: 20.855426788330078 ms per batch image class_id:0, confidence:0.91,left_top:[859.85,209.22], right_bottom:[921.85,370.21] class_id:17, confidence:0.99,left_top:[736.66,251.05], right_bottom:[1005.21,438.58] class_id:17, confidence:0.96,left_top:[140.42,274.80], right_bottom:[385.45,467.28] class_id:17, confidence:0.90,left_top:[472.41,285.53], right_bottom:[695.09,448.15] class_id:17, confidence:0.36,left_top:[585.75,283.86], right_bottom:[717.58,445.50] class_id:17, confidence:0.27,left_top:[0.00,315.37], right_bottom:[40.32,380.57] class_id:17, confidence:0.27,left_top:[0.53,304.21], right_bottom:[43.35,474.99] detect frame:193 Inference: 21.04640007019043 ms per batch image class_id:0, confidence:0.85,left_top:[856.64,205.43], right_bottom:[919.64,368.03] class_id:17, confidence:0.99,left_top:[735.47,248.05], right_bottom:[1001.80,440.76] class_id:17, confidence:0.96,left_top:[464.42,285.63], right_bottom:[703.51,449.89] class_id:17, confidence:0.70,left_top:[137.55,274.81], right_bottom:[378.55,467.22] class_id:17, confidence:0.35,left_top:[0.00,316.19], right_bottom:[39.21,382.95] detect frame:194 Inference: 20.919322967529297 ms per batch image class_id:0, confidence:0.68,left_top:[848.12,213.24], right_bottom:[909.20,360.15] class_id:17, confidence:0.99,left_top:[728.97,239.72], right_bottom:[997.29,435.25] class_id:17, confidence:0.97,left_top:[474.19,291.32], right_bottom:[697.47,454.02] class_id:17, confidence:0.84,left_top:[211.52,295.76], right_bottom:[377.34,470.99] class_id:17, confidence:0.46,left_top:[449.84,334.65], right_bottom:[582.10,443.56] class_id:17, confidence:0.34,left_top:[130.95,343.10], right_bottom:[236.42,474.89] detect frame:195 Inference: 20.92432975769043 ms per batch image class_id:0, confidence:0.91,left_top:[835.31,204.96], right_bottom:[901.18,353.08] class_id:17, confidence:0.99,left_top:[459.07,288.61], right_bottom:[679.41,457.09] class_id:17, confidence:0.99,left_top:[723.85,233.10], right_bottom:[987.41,430.63] class_id:17, confidence:0.83,left_top:[204.95,298.45], right_bottom:[366.48,470.84] detect frame:196 Inference: 20.837783813476562 ms per batch image class_id:0, confidence:0.94,left_top:[834.97,194.95], right_bottom:[903.00,353.15] class_id:17, confidence:0.99,left_top:[718.39,223.99], right_bottom:[988.43,427.21] class_id:17, confidence:0.99,left_top:[459.71,284.42], right_bottom:[680.01,450.45] class_id:17, confidence:0.90,left_top:[121.02,293.68], right_bottom:[367.10,480.84] detect frame:197 Inference: 20.696401596069336 ms per batch image class_id:0, confidence:0.84,left_top:[834.60,209.22], right_bottom:[901.51,366.39] class_id:17, confidence:0.99,left_top:[725.24,237.61], right_bottom:[988.62,435.75] class_id:17, confidence:0.98,left_top:[104.69,293.49], right_bottom:[364.79,480.28] class_id:17, confidence:0.97,left_top:[456.27,293.08], right_bottom:[681.67,449.81] detect frame:198 Inference: 20.77770233154297 ms per batch image class_id:0, confidence:0.76,left_top:[839.72,202.14], right_bottom:[903.29,360.45] class_id:17, confidence:0.99,left_top:[725.07,232.39], right_bottom:[988.47,436.06] class_id:17, confidence:0.95,left_top:[456.37,280.77], right_bottom:[678.89,444.46] class_id:17, confidence:0.93,left_top:[136.94,283.46], right_bottom:[364.96,464.74] detect frame:199 Inference: 20.816564559936523 ms per batch image class_id:0, confidence:0.70,left_top:[836.74,200.50], right_bottom:[903.21,360.09] class_id:17, confidence:0.99,left_top:[729.97,228.12], right_bottom:[988.03,432.06] class_id:17, confidence:0.93,left_top:[456.75,275.65], right_bottom:[677.53,442.99] class_id:17, confidence:0.87,left_top:[130.37,284.21], right_bottom:[359.51,463.72] detect frame:200 Inference: 20.893096923828125 ms per batch image class_id:0, confidence:0.73,left_top:[831.17,207.82], right_bottom:[897.72,367.76] class_id:17, confidence:0.99,left_top:[726.97,231.81], right_bottom:[984.58,436.92] class_id:17, confidence:0.91,left_top:[460.84,277.84], right_bottom:[670.29,441.57] class_id:17, confidence:0.91,left_top:[121.87,290.71], right_bottom:[360.06,458.73] class_id:33, confidence:0.23,left_top:[984.03,28.17], right_bottom:[1054.08,106.51] detect frame:201 Inference: 20.65420150756836 ms per batch image class_id:0, confidence:0.73,left_top:[834.35,205.93], right_bottom:[899.80,357.06] class_id:17, confidence:0.99,left_top:[729.61,235.60], right_bottom:[981.86,431.93] class_id:17, confidence:0.91,left_top:[116.61,298.85], right_bottom:[353.63,466.45] class_id:17, confidence:0.84,left_top:[438.51,281.61], right_bottom:[620.30,426.82] class_id:17, confidence:0.81,left_top:[560.80,275.11], right_bottom:[687.49,437.13] detect frame:202 Inference: 20.571470260620117 ms per batch image class_id:0, confidence:0.83,left_top:[833.25,203.52], right_bottom:[901.64,356.04] class_id:17, confidence:0.99,left_top:[719.79,227.81], right_bottom:[984.83,426.95] class_id:17, confidence:0.97,left_top:[459.72,285.64], right_bottom:[683.99,447.19] class_id:17, confidence:0.84,left_top:[123.61,298.01], right_bottom:[349.68,469.37] detect frame:203 Inference: 20.741939544677734 ms per batch image class_id:0, confidence:0.84,left_top:[839.26,191.81], right_bottom:[901.49,350.38] class_id:17, confidence:0.99,left_top:[726.66,219.57], right_bottom:[986.25,419.06] class_id:17, confidence:0.89,left_top:[123.32,288.25], right_bottom:[362.21,460.42] class_id:17, confidence:0.81,left_top:[469.69,276.15], right_bottom:[680.67,441.90] class_id:17, confidence:0.34,left_top:[100.74,319.57], right_bottom:[258.83,455.27] detect frame:204 Inference: 20.821332931518555 ms per batch image class_id:0, confidence:0.85,left_top:[838.11,181.39], right_bottom:[902.73,338.58] class_id:17, confidence:0.99,left_top:[724.29,214.11], right_bottom:[986.10,417.58] class_id:17, confidence:0.82,left_top:[118.29,286.09], right_bottom:[359.14,458.32] class_id:17, confidence:0.72,left_top:[467.40,272.50], right_bottom:[675.55,440.64] class_id:17, confidence:0.27,left_top:[99.66,317.67], right_bottom:[256.78,454.43] detect frame:205 Inference: 20.609140396118164 ms per batch image class_id:0, confidence:0.93,left_top:[847.36,171.98], right_bottom:[909.82,339.86] class_id:17, confidence:0.99,left_top:[726.35,209.26], right_bottom:[992.23,415.04] class_id:17, confidence:0.91,left_top:[94.57,278.25], right_bottom:[367.74,457.30] class_id:17, confidence:0.77,left_top:[466.59,270.31], right_bottom:[677.19,438.78] detect frame:206 Inference: 20.828723907470703 ms per batch image class_id:0, confidence:0.88,left_top:[853.67,178.18], right_bottom:[914.70,346.24] class_id:17, confidence:0.99,left_top:[736.68,215.62], right_bottom:[997.12,417.56] class_id:17, confidence:0.96,left_top:[116.58,290.73], right_bottom:[368.25,451.76] class_id:17, confidence:0.75,left_top:[470.99,275.95], right_bottom:[687.97,437.06] detect frame:207 Inference: 20.7366943359375 ms per batch image class_id:0, confidence:0.84,left_top:[850.74,166.48], right_bottom:[910.87,339.53] class_id:17, confidence:0.99,left_top:[738.92,205.40], right_bottom:[993.43,409.06] class_id:17, confidence:0.90,left_top:[458.14,269.38], right_bottom:[683.92,433.57] class_id:17, confidence:0.82,left_top:[129.28,283.65], right_bottom:[362.70,451.27] class_id:17, confidence:0.28,left_top:[433.66,314.70], right_bottom:[558.84,422.80] detect frame:208 Inference: 20.885944366455078 ms per batch image class_id:0, confidence:0.80,left_top:[851.07,163.40], right_bottom:[907.98,337.93] class_id:17, confidence:0.98,left_top:[725.53,207.29], right_bottom:[994.62,405.72] class_id:17, confidence:0.91,left_top:[454.06,264.93], right_bottom:[681.37,431.88] class_id:17, confidence:0.89,left_top:[131.98,280.33], right_bottom:[367.02,453.13] class_id:17, confidence:0.31,left_top:[109.43,322.98], right_bottom:[204.95,450.47] detect frame:209 Inference: 20.88785171508789 ms per batch image class_id:0, confidence:0.77,left_top:[854.38,183.46], right_bottom:[904.40,340.12] class_id:17, confidence:0.99,left_top:[728.45,217.01], right_bottom:[989.02,413.64] class_id:17, confidence:0.94,left_top:[458.74,278.17], right_bottom:[681.37,430.98] class_id:17, confidence:0.91,left_top:[121.99,288.30], right_bottom:[373.44,451.02] class_id:17, confidence:0.28,left_top:[103.33,327.88], right_bottom:[211.65,450.60] detect frame:210 Inference: 20.832300186157227 ms per batch image class_id:0, confidence:0.94,left_top:[858.80,168.13], right_bottom:[909.68,334.39] class_id:17, confidence:0.99,left_top:[732.08,209.30], right_bottom:[995.53,405.32] class_id:17, confidence:0.86,left_top:[472.95,256.94], right_bottom:[694.55,416.43] class_id:17, confidence:0.82,left_top:[117.80,265.80], right_bottom:[379.18,445.16] detect frame:211 Inference: 20.719051361083984 ms per batch image class_id:0, confidence:0.90,left_top:[861.19,166.18], right_bottom:[912.47,337.05] class_id:17, confidence:0.99,left_top:[730.70,211.73], right_bottom:[997.91,405.68] class_id:17, confidence:0.96,left_top:[471.04,256.65], right_bottom:[700.51,418.11] class_id:17, confidence:0.65,left_top:[208.28,263.02], right_bottom:[378.16,435.11] class_id:17, confidence:0.24,left_top:[103.95,301.77], right_bottom:[261.05,437.45] detect frame:212 Inference: 20.73049545288086 ms per batch image class_id:0, confidence:0.92,left_top:[855.67,155.46], right_bottom:[916.25,318.45] class_id:17, confidence:0.99,left_top:[723.07,197.71], right_bottom:[1004.94,395.54] class_id:17, confidence:0.97,left_top:[471.38,253.16], right_bottom:[705.27,418.15] class_id:17, confidence:0.84,left_top:[229.18,252.61], right_bottom:[392.31,419.69] class_id:17, confidence:0.61,left_top:[111.53,273.97], right_bottom:[271.30,432.15] detect frame:213 Inference: 21.02351188659668 ms per batch image class_id:0, confidence:0.85,left_top:[858.54,169.47], right_bottom:[920.42,329.93] class_id:12, confidence:0.30,left_top:[1191.32,449.37], right_bottom:[1245.41,603.51] class_id:17, confidence:0.99,left_top:[728.38,200.18], right_bottom:[1006.46,394.90] class_id:17, confidence:0.98,left_top:[479.73,260.23], right_bottom:[707.78,413.91] class_id:17, confidence:0.87,left_top:[226.19,258.91], right_bottom:[395.39,415.61] class_id:17, confidence:0.56,left_top:[116.49,305.35], right_bottom:[248.25,428.72] class_id:67, confidence:0.63,left_top:[1191.32,449.37], right_bottom:[1245.41,603.51] detect frame:214 Inference: 20.769357681274414 ms per batch image class_id:0, confidence:0.89,left_top:[863.05,166.42], right_bottom:[928.37,320.87] class_id:17, confidence:0.99,left_top:[731.68,201.24], right_bottom:[1009.16,394.95] class_id:17, confidence:0.87,left_top:[486.92,252.96], right_bottom:[698.32,419.67] class_id:17, confidence:0.61,left_top:[231.66,264.01], right_bottom:[397.74,431.49] class_id:17, confidence:0.40,left_top:[130.36,276.04], right_bottom:[350.35,427.61] detect frame:215 Inference: 20.819425582885742 ms per batch image class_id:0, confidence:0.90,left_top:[871.22,161.59], right_bottom:[938.50,315.55] class_id:17, confidence:0.99,left_top:[743.91,198.17], right_bottom:[1022.22,391.76] class_id:17, confidence:0.73,left_top:[182.80,251.16], right_bottom:[398.76,411.51] class_id:17, confidence:0.60,left_top:[486.01,247.88], right_bottom:[696.53,411.84] class_id:17, confidence:0.23,left_top:[140.66,289.07], right_bottom:[329.71,407.87] detect frame:216 Inference: 20.859956741333008 ms per batch image class_id:0, confidence:0.81,left_top:[880.75,163.97], right_bottom:[946.33,321.24] class_id:17, confidence:0.99,left_top:[743.08,195.95], right_bottom:[1031.24,395.83] class_id:17, confidence:0.80,left_top:[203.20,246.62], right_bottom:[410.80,412.67] class_id:17, confidence:0.77,left_top:[523.52,249.48], right_bottom:[732.91,413.50] class_id:17, confidence:0.23,left_top:[138.91,294.12], right_bottom:[246.58,401.84] detect frame:217 Inference: 20.86353302001953 ms per batch image class_id:0, confidence:0.82,left_top:[887.44,161.65], right_bottom:[949.68,326.88] class_id:12, confidence:0.54,left_top:[1031.00,454.04], right_bottom:[1075.34,617.73] class_id:17, confidence:0.98,left_top:[741.46,196.03], right_bottom:[1041.50,395.34] class_id:17, confidence:0.81,left_top:[531.25,243.18], right_bottom:[739.08,409.78] class_id:17, confidence:0.76,left_top:[262.10,245.87], right_bottom:[436.76,407.89] class_id:17, confidence:0.36,left_top:[148.54,289.28], right_bottom:[284.47,404.40] class_id:17, confidence:0.20,left_top:[0.80,253.34], right_bottom:[49.77,406.61] detect frame:218 Inference: 20.786046981811523 ms per batch image class_id:0, confidence:0.91,left_top:[889.06,166.04], right_bottom:[952.04,331.13] class_id:12, confidence:0.52,left_top:[996.44,455.92], right_bottom:[1042.42,618.30] class_id:17, confidence:0.99,left_top:[738.81,195.20], right_bottom:[1043.68,395.78] class_id:17, confidence:0.88,left_top:[531.09,238.35], right_bottom:[739.76,406.37] class_id:17, confidence:0.71,left_top:[249.95,242.31], right_bottom:[448.66,401.16] class_id:17, confidence:0.51,left_top:[0.16,246.33], right_bottom:[60.78,410.51] class_id:17, confidence:0.27,left_top:[156.96,293.49], right_bottom:[286.50,402.13] class_id:67, confidence:0.39,left_top:[996.44,455.92], right_bottom:[1042.42,618.30] detect frame:219 Inference: 20.817041397094727 ms per batch image class_id:0, confidence:0.92,left_top:[889.57,166.85], right_bottom:[956.34,331.39] class_id:12, confidence:0.35,left_top:[961.79,455.93], right_bottom:[1008.07,618.65] class_id:17, confidence:0.98,left_top:[751.95,193.51], right_bottom:[1048.17,393.51] class_id:17, confidence:0.79,left_top:[539.35,237.08], right_bottom:[745.72,402.34] class_id:17, confidence:0.76,left_top:[254.52,244.03], right_bottom:[454.28,408.47] class_id:17, confidence:0.69,left_top:[2.48,244.01], right_bottom:[70.56,407.60] class_id:17, confidence:0.34,left_top:[170.58,294.93], right_bottom:[280.08,399.34] class_id:17, confidence:0.22,left_top:[205.91,252.61], right_bottom:[360.07,405.68] detect frame:220 Inference: 20.712852478027344 ms per batch image class_id:0, confidence:0.85,left_top:[898.83,166.40], right_bottom:[966.21,317.17] class_id:12, confidence:0.31,left_top:[927.78,458.62], right_bottom:[972.31,615.06] class_id:17, confidence:0.99,left_top:[548.92,237.26], right_bottom:[765.57,404.69] class_id:17, confidence:0.98,left_top:[759.97,192.92], right_bottom:[1054.30,396.72] class_id:17, confidence:0.76,left_top:[0.00,241.74], right_bottom:[77.49,395.82] class_id:17, confidence:0.59,left_top:[303.53,247.02], right_bottom:[464.25,412.41] class_id:17, confidence:0.50,left_top:[186.54,248.64], right_bottom:[393.77,403.88] detect frame:221 Inference: 20.910024642944336 ms per batch image class_id:0, confidence:0.84,left_top:[903.78,162.05], right_bottom:[971.21,316.97] class_id:12, confidence:0.21,left_top:[896.99,455.16], right_bottom:[939.26,620.17] class_id:17, confidence:0.99,left_top:[767.93,192.65], right_bottom:[1063.78,391.02] class_id:17, confidence:0.98,left_top:[551.79,240.07], right_bottom:[768.95,402.83] class_id:17, confidence:0.87,left_top:[271.98,248.45], right_bottom:[476.45,407.75] class_id:17, confidence:0.80,left_top:[188.87,249.63], right_bottom:[396.31,397.32] class_id:17, confidence:0.79,left_top:[0.00,242.43], right_bottom:[77.54,406.41] detect frame:222 Inference: 20.99609375 ms per batch image class_id:0, confidence:0.89,left_top:[922.30,151.06], right_bottom:[981.28,319.31] class_id:17, confidence:0.99,left_top:[778.33,185.04], right_bottom:[1075.61,390.45] class_id:17, confidence:0.94,left_top:[198.58,244.47], right_bottom:[402.10,396.69] class_id:17, confidence:0.93,left_top:[306.38,248.22], right_bottom:[514.59,412.44] class_id:17, confidence:0.92,left_top:[0.00,238.98], right_bottom:[94.24,397.96] class_id:17, confidence:0.82,left_top:[547.78,246.23], right_bottom:[772.32,396.64] detect frame:223 Inference: 20.842552185058594 ms per batch image class_id:0, confidence:0.93,left_top:[926.24,148.37], right_bottom:[989.53,319.01] class_id:17, confidence:0.99,left_top:[787.29,188.30], right_bottom:[1085.16,390.89] class_id:17, confidence:0.94,left_top:[0.00,237.45], right_bottom:[98.39,393.14] class_id:17, confidence:0.93,left_top:[581.54,246.70], right_bottom:[787.47,398.64] class_id:17, confidence:0.81,left_top:[311.58,249.33], right_bottom:[518.40,399.51] class_id:17, confidence:0.52,left_top:[205.71,254.88], right_bottom:[403.34,394.74] detect frame:224 Inference: 20.889997482299805 ms per batch image class_id:0, confidence:0.86,left_top:[933.67,147.40], right_bottom:[1000.74,315.19] class_id:17, confidence:0.98,left_top:[795.91,183.70], right_bottom:[1099.93,391.56] class_id:17, confidence:0.97,left_top:[583.84,242.86], right_bottom:[793.39,396.79] class_id:17, confidence:0.94,left_top:[0.00,235.80], right_bottom:[99.74,397.89] class_id:17, confidence:0.93,left_top:[313.27,248.27], right_bottom:[516.81,406.23] class_id:17, confidence:0.50,left_top:[208.23,255.68], right_bottom:[405.84,396.75] detect frame:225 Inference: 20.67732810974121 ms per batch image class_id:0, confidence:0.88,left_top:[938.08,145.35], right_bottom:[1004.42,315.82] class_id:12, confidence:0.20,left_top:[727.86,457.38], right_bottom:[766.89,616.81] class_id:17, confidence:0.99,left_top:[797.80,181.89], right_bottom:[1104.35,387.50] class_id:17, confidence:0.91,left_top:[1.34,234.86], right_bottom:[106.88,402.28] class_id:17, confidence:0.91,left_top:[317.82,248.43], right_bottom:[520.84,406.98] class_id:17, confidence:0.82,left_top:[526.86,241.68], right_bottom:[788.27,390.22] class_id:17, confidence:0.46,left_top:[224.11,278.23], right_bottom:[341.20,384.92] detect frame:226 Inference: 20.792007446289062 ms per batch image class_id:0, confidence:0.78,left_top:[948.42,142.47], right_bottom:[1011.25,306.38] class_id:17, confidence:0.99,left_top:[819.09,180.47], right_bottom:[1105.01,384.79] class_id:17, confidence:0.95,left_top:[588.10,231.08], right_bottom:[802.52,394.19] class_id:17, confidence:0.91,left_top:[0.00,234.59], right_bottom:[116.51,399.92] class_id:17, confidence:0.77,left_top:[323.05,248.95], right_bottom:[528.81,402.95] class_id:17, confidence:0.32,left_top:[233.36,263.77], right_bottom:[391.50,393.28] detect frame:227 Inference: 20.930051803588867 ms per batch image class_id:0, confidence:0.82,left_top:[957.07,143.04], right_bottom:[1013.68,316.63] class_id:17, confidence:0.99,left_top:[846.37,182.77], right_bottom:[1110.84,384.14] class_id:17, confidence:0.97,left_top:[611.95,222.10], right_bottom:[815.30,395.75] class_id:17, confidence:0.88,left_top:[4.04,241.90], right_bottom:[138.14,395.49] class_id:17, confidence:0.84,left_top:[340.42,245.64], right_bottom:[542.62,399.15] class_id:17, confidence:0.41,left_top:[256.03,261.71], right_bottom:[457.30,395.87] detect frame:228 Inference: 20.78986167907715 ms per batch image class_id:0, confidence:0.87,left_top:[960.24,143.12], right_bottom:[1022.40,318.11] class_id:12, confidence:0.26,left_top:[598.56,457.32], right_bottom:[637.13,613.26] class_id:17, confidence:0.99,left_top:[851.37,180.07], right_bottom:[1116.12,384.92] class_id:17, confidence:0.97,left_top:[617.72,226.06], right_bottom:[823.99,393.56] class_id:17, confidence:0.95,left_top:[338.63,244.75], right_bottom:[543.63,406.62] class_id:17, confidence:0.93,left_top:[1.52,248.02], right_bottom:[139.06,405.04] class_id:17, confidence:0.55,left_top:[250.12,273.90], right_bottom:[400.64,391.65] detect frame:229 Inference: 20.82061767578125 ms per batch image class_id:0, confidence:0.94,left_top:[965.49,145.08], right_bottom:[1029.45,313.90] class_id:17, confidence:0.99,left_top:[854.16,179.72], right_bottom:[1118.79,380.60] class_id:17, confidence:0.97,left_top:[618.82,226.83], right_bottom:[826.87,397.54] class_id:17, confidence:0.96,left_top:[1.10,249.96], right_bottom:[139.85,409.85] class_id:17, confidence:0.83,left_top:[338.89,244.05], right_bottom:[544.79,405.39] class_id:17, confidence:0.57,left_top:[259.96,277.22], right_bottom:[393.68,392.85] detect frame:230 Inference: 20.766019821166992 ms per batch image class_id:0, confidence:0.85,left_top:[971.65,141.00], right_bottom:[1034.53,309.40] class_id:12, confidence:0.64,left_top:[533.43,465.06], right_bottom:[574.60,609.19] class_id:17, confidence:0.99,left_top:[849.62,181.71], right_bottom:[1125.15,385.44] class_id:17, confidence:0.98,left_top:[620.75,224.32], right_bottom:[834.94,399.73] class_id:17, confidence:0.96,left_top:[1.38,254.53], right_bottom:[145.91,411.49] class_id:17, confidence:0.93,left_top:[346.78,250.41], right_bottom:[559.48,415.24] class_id:17, confidence:0.39,left_top:[258.02,257.07], right_bottom:[464.83,403.66] detect frame:231 Inference: 20.79463005065918 ms per batch image class_id:0, confidence:0.84,left_top:[970.67,140.88], right_bottom:[1038.97,306.14] class_id:12, confidence:0.21,left_top:[503.14,466.50], right_bottom:[543.47,612.68] class_id:17, confidence:0.99,left_top:[850.21,180.89], right_bottom:[1132.81,381.37] class_id:17, confidence:0.98,left_top:[614.06,227.44], right_bottom:[839.05,399.52] class_id:17, confidence:0.98,left_top:[0.00,261.58], right_bottom:[150.99,411.56] class_id:17, confidence:0.85,left_top:[405.80,247.30], right_bottom:[568.64,415.50] class_id:17, confidence:0.62,left_top:[261.79,260.50], right_bottom:[430.53,404.25] detect frame:232 Inference: 20.9810733795166 ms per batch image class_id:0, confidence:0.89,left_top:[982.16,142.67], right_bottom:[1049.87,307.88] class_id:17, confidence:0.99,left_top:[862.62,181.93], right_bottom:[1143.48,386.23] class_id:17, confidence:0.99,left_top:[0.00,277.57], right_bottom:[160.44,418.87] class_id:17, confidence:0.97,left_top:[623.05,239.39], right_bottom:[847.56,405.05] class_id:17, confidence:0.79,left_top:[405.86,256.48], right_bottom:[578.72,417.13] class_id:17, confidence:0.28,left_top:[276.14,278.48], right_bottom:[426.39,417.52] detect frame:233 Inference: 20.91360092163086 ms per batch image class_id:0, confidence:0.94,left_top:[988.05,140.50], right_bottom:[1052.56,309.87] class_id:17, confidence:0.99,left_top:[858.25,178.64], right_bottom:[1151.56,386.53] class_id:17, confidence:0.99,left_top:[0.00,275.17], right_bottom:[164.03,415.73] class_id:17, confidence:0.97,left_top:[620.28,239.59], right_bottom:[848.60,404.28] class_id:17, confidence:0.82,left_top:[434.72,254.27], right_bottom:[590.44,417.20] class_id:17, confidence:0.38,left_top:[298.87,267.84], right_bottom:[469.15,423.73] class_id:17, confidence:0.21,left_top:[285.87,310.66], right_bottom:[364.54,399.23] detect frame:234 Inference: 20.9352970123291 ms per batch image class_id:0, confidence:0.86,left_top:[992.05,145.03], right_bottom:[1057.64,317.55] class_id:12, confidence:0.34,left_top:[377.23,482.62], right_bottom:[415.16,629.50] class_id:17, confidence:0.99,left_top:[647.02,243.92], right_bottom:[857.40,404.12] class_id:17, confidence:0.99,left_top:[869.73,180.91], right_bottom:[1163.53,394.69] class_id:17, confidence:0.99,left_top:[0.00,273.28], right_bottom:[174.15,415.37] class_id:17, confidence:0.76,left_top:[436.02,267.03], right_bottom:[605.50,427.22] class_id:17, confidence:0.32,left_top:[379.04,266.91], right_bottom:[522.62,426.17] detect frame:235 Inference: 20.863771438598633 ms per batch image class_id:0, confidence:0.92,left_top:[993.73,154.90], right_bottom:[1057.60,328.07] class_id:12, confidence:0.23,left_top:[344.19,487.18], right_bottom:[383.43,632.01] class_id:17, confidence:0.99,left_top:[0.00,270.41], right_bottom:[181.09,420.48] class_id:17, confidence:0.99,left_top:[872.49,189.71], right_bottom:[1162.04,402.00] class_id:17, confidence:0.99,left_top:[654.65,249.57], right_bottom:[867.53,407.89] class_id:17, confidence:0.87,left_top:[443.54,264.82], right_bottom:[612.71,428.13] class_id:17, confidence:0.40,left_top:[319.46,273.71], right_bottom:[525.08,422.56] class_id:17, confidence:0.35,left_top:[404.88,271.09], right_bottom:[515.69,320.39] detect frame:236 Inference: 20.882606506347656 ms per batch image class_id:0, confidence:0.93,left_top:[994.04,165.43], right_bottom:[1057.39,335.15] class_id:12, confidence:0.21,left_top:[313.23,492.95], right_bottom:[351.89,634.49] class_id:17, confidence:0.99,left_top:[882.39,205.30], right_bottom:[1160.49,407.63] class_id:17, confidence:0.99,left_top:[657.52,254.91], right_bottom:[867.90,409.48] class_id:17, confidence:0.99,left_top:[0.00,276.11], right_bottom:[185.66,420.87] class_id:17, confidence:0.86,left_top:[437.02,270.39], right_bottom:[621.14,426.52] class_id:17, confidence:0.79,left_top:[316.82,274.31], right_bottom:[525.39,428.40] detect frame:237 Inference: 20.967483520507812 ms per batch image class_id:0, confidence:0.88,left_top:[997.31,181.36], right_bottom:[1062.15,343.28] class_id:17, confidence:0.99,left_top:[883.83,211.69], right_bottom:[1166.11,416.23] class_id:17, confidence:0.98,left_top:[3.14,278.28], right_bottom:[181.43,428.86] class_id:17, confidence:0.98,left_top:[675.76,257.93], right_bottom:[877.51,415.41] class_id:17, confidence:0.94,left_top:[422.67,272.69], right_bottom:[632.71,435.81] class_id:17, confidence:0.82,left_top:[306.43,277.02], right_bottom:[480.90,419.67] detect frame:238 Inference: 20.914316177368164 ms per batch image class_id:0, confidence:0.75,left_top:[995.30,201.48], right_bottom:[1062.50,357.54] class_id:17, confidence:0.99,left_top:[875.38,228.58], right_bottom:[1163.19,431.92] class_id:17, confidence:0.98,left_top:[677.52,270.46], right_bottom:[876.44,424.34] class_id:17, confidence:0.97,left_top:[2.04,280.61], right_bottom:[181.99,431.21] class_id:17, confidence:0.86,left_top:[298.95,284.86], right_bottom:[487.02,424.33] class_id:17, confidence:0.86,left_top:[462.92,274.96], right_bottom:[636.61,430.21] detect frame:239 Inference: 20.718812942504883 ms per batch image class_id:0, confidence:0.68,left_top:[995.45,204.62], right_bottom:[1060.56,356.86] class_id:12, confidence:0.22,left_top:[173.97,500.93], right_bottom:[207.87,647.27] class_id:17, confidence:0.99,left_top:[878.71,233.16], right_bottom:[1159.91,437.49] class_id:17, confidence:0.98,left_top:[687.74,273.34], right_bottom:[879.57,421.42] class_id:17, confidence:0.96,left_top:[2.93,282.39], right_bottom:[184.39,431.86] class_id:17, confidence:0.91,left_top:[444.01,280.03], right_bottom:[645.57,438.32] class_id:17, confidence:0.82,left_top:[302.67,284.20], right_bottom:[484.08,426.25] detect frame:240 Inference: 20.952224731445312 ms per batch image class_id:0, confidence:0.85,left_top:[997.02,196.10], right_bottom:[1062.68,357.49] class_id:17, confidence:0.99,left_top:[877.31,231.37], right_bottom:[1164.01,432.09] class_id:17, confidence:0.99,left_top:[688.70,262.17], right_bottom:[883.66,434.82] class_id:17, confidence:0.97,left_top:[4.02,282.05], right_bottom:[185.37,430.66] class_id:17, confidence:0.95,left_top:[309.42,278.09], right_bottom:[485.28,425.78] class_id:17, confidence:0.94,left_top:[443.71,269.22], right_bottom:[643.95,435.67] detect frame:241 Inference: 20.900726318359375 ms per batch image class_id:0, confidence:0.90,left_top:[997.05,212.12], right_bottom:[1062.95,369.77] class_id:17, confidence:0.99,left_top:[874.19,238.46], right_bottom:[1160.36,437.33] class_id:17, confidence:0.99,left_top:[2.01,302.14], right_bottom:[182.41,440.32] class_id:17, confidence:0.98,left_top:[685.96,282.74], right_bottom:[879.50,438.45] class_id:17, confidence:0.86,left_top:[445.37,285.94], right_bottom:[646.18,440.69] class_id:17, confidence:0.83,left_top:[300.00,295.94], right_bottom:[498.33,439.32] detect frame:242 Inference: 20.9197998046875 ms per batch image class_id:0, confidence:0.93,left_top:[988.04,189.56], right_bottom:[1048.93,355.17] class_id:17, confidence:0.99,left_top:[863.05,226.85], right_bottom:[1145.62,436.90] class_id:17, confidence:0.98,left_top:[0.00,301.81], right_bottom:[181.80,449.59] class_id:17, confidence:0.98,left_top:[674.69,269.37], right_bottom:[870.13,444.57] class_id:17, confidence:0.93,left_top:[463.87,278.65], right_bottom:[632.13,450.33] class_id:17, confidence:0.90,left_top:[315.38,294.78], right_bottom:[481.34,442.22] detect frame:243 Inference: 20.920515060424805 ms per batch image class_id:0, confidence:0.86,left_top:[986.52,203.16], right_bottom:[1046.53,370.32] class_id:12, confidence:0.23,left_top:[3.55,530.37], right_bottom:[36.28,669.60] class_id:17, confidence:0.99,left_top:[859.19,245.79], right_bottom:[1143.13,445.60] class_id:17, confidence:0.99,left_top:[0.74,311.57], right_bottom:[184.19,452.47] class_id:17, confidence:0.97,left_top:[671.02,283.18], right_bottom:[866.18,444.34] class_id:17, confidence:0.79,left_top:[469.46,284.55], right_bottom:[632.37,450.07] class_id:17, confidence:0.62,left_top:[338.68,303.32], right_bottom:[485.36,447.36] detect frame:244 Inference: 20.93338966369629 ms per batch image class_id:0, confidence:0.92,left_top:[986.79,201.81], right_bottom:[1046.86,373.27] class_id:17, confidence:0.99,left_top:[862.19,244.80], right_bottom:[1142.70,446.74] class_id:17, confidence:0.99,left_top:[0.00,301.04], right_bottom:[188.55,447.76] class_id:17, confidence:0.97,left_top:[670.34,281.46], right_bottom:[866.81,442.61] class_id:17, confidence:0.86,left_top:[470.31,284.47], right_bottom:[633.83,454.72] class_id:17, confidence:0.67,left_top:[317.49,301.38], right_bottom:[473.34,444.63] detect frame:245 Inference: 21.015167236328125 ms per batch image class_id:0, confidence:0.91,left_top:[974.07,219.56], right_bottom:[1038.38,380.12] class_id:17, confidence:0.99,left_top:[852.66,252.52], right_bottom:[1133.56,450.46] class_id:17, confidence:0.99,left_top:[664.49,293.37], right_bottom:[863.52,447.03] class_id:17, confidence:0.98,left_top:[0.00,320.29], right_bottom:[188.24,462.44] class_id:17, confidence:0.92,left_top:[424.19,303.58], right_bottom:[629.79,467.73] class_id:17, confidence:0.86,left_top:[312.58,319.14], right_bottom:[481.31,460.46] detect frame:246 Inference: 20.99752426147461 ms per batch image class_id:0, confidence:0.86,left_top:[966.18,211.92], right_bottom:[1026.21,378.32] class_id:17, confidence:0.99,left_top:[843.28,245.51], right_bottom:[1134.72,443.58] class_id:17, confidence:0.99,left_top:[650.66,286.89], right_bottom:[849.40,449.90] class_id:17, confidence:0.98,left_top:[2.39,337.84], right_bottom:[172.91,479.20] class_id:17, confidence:0.90,left_top:[414.13,309.39], right_bottom:[614.45,471.27] class_id:17, confidence:0.70,left_top:[303.28,324.48], right_bottom:[466.32,464.97] detect frame:247 Inference: 20.88189125061035 ms per batch image class_id:0, confidence:0.82,left_top:[970.15,227.10], right_bottom:[1031.02,395.53] class_id:17, confidence:0.99,left_top:[1.90,337.07], right_bottom:[175.40,479.93] class_id:17, confidence:0.99,left_top:[842.03,263.11], right_bottom:[1132.33,462.71] class_id:17, confidence:0.97,left_top:[647.40,303.40], right_bottom:[853.49,465.56] class_id:17, confidence:0.75,left_top:[473.13,314.36], right_bottom:[630.69,465.58] class_id:17, confidence:0.37,left_top:[391.93,328.28], right_bottom:[522.23,480.10] class_id:17, confidence:0.35,left_top:[315.10,326.09], right_bottom:[463.11,460.13] detect frame:248 Inference: 20.854949951171875 ms per batch image class_id:0, confidence:0.85,left_top:[968.37,243.42], right_bottom:[1030.50,406.32] class_id:17, confidence:0.99,left_top:[849.93,272.60], right_bottom:[1126.31,469.73] class_id:17, confidence:0.99,left_top:[0.02,339.39], right_bottom:[172.09,478.91] class_id:17, confidence:0.94,left_top:[650.34,317.29], right_bottom:[853.09,464.47] class_id:17, confidence:0.84,left_top:[471.40,330.27], right_bottom:[637.18,473.50] class_id:17, confidence:0.56,left_top:[282.10,345.25], right_bottom:[492.96,473.04] class_id:17, confidence:0.45,left_top:[390.72,332.34], right_bottom:[529.32,482.71] detect frame:249 Inference: 21.005868911743164 ms per batch image class_id:0, confidence:0.86,left_top:[964.78,234.53], right_bottom:[1027.00,402.52] class_id:17, confidence:0.99,left_top:[836.24,267.34], right_bottom:[1125.85,467.11] class_id:17, confidence:0.99,left_top:[2.11,333.94], right_bottom:[169.56,477.39] class_id:17, confidence:0.89,left_top:[415.77,331.91], right_bottom:[632.59,475.55] class_id:17, confidence:0.88,left_top:[646.08,315.92], right_bottom:[852.07,469.46] class_id:17, confidence:0.38,left_top:[313.58,334.14], right_bottom:[531.28,474.50] detect frame:250 Inference: 21.170377731323242 ms per batch image class_id:0, confidence:0.93,left_top:[961.05,231.13], right_bottom:[1024.79,404.31] class_id:17, confidence:0.99,left_top:[4.08,334.09], right_bottom:[168.31,478.76] class_id:17, confidence:0.98,left_top:[835.08,264.49], right_bottom:[1121.38,472.39] class_id:17, confidence:0.95,left_top:[417.67,335.15], right_bottom:[634.02,479.68] class_id:17, confidence:0.95,left_top:[643.20,314.21], right_bottom:[850.97,470.01] class_id:17, confidence:0.46,left_top:[322.94,334.14], right_bottom:[528.54,474.93] detect frame:251 Inference: 20.951509475708008 ms per batch image class_id:0, confidence:0.81,left_top:[959.85,240.40], right_bottom:[1025.04,409.38] class_id:17, confidence:0.99,left_top:[837.95,270.61], right_bottom:[1124.49,478.04] class_id:17, confidence:0.98,left_top:[0.00,341.47], right_bottom:[173.62,482.92] class_id:17, confidence:0.97,left_top:[645.77,322.17], right_bottom:[850.59,475.83] class_id:17, confidence:0.69,left_top:[418.70,338.04], right_bottom:[635.74,485.94] class_id:17, confidence:0.57,left_top:[319.48,341.47], right_bottom:[467.72,477.73] detect frame:252 Inference: 20.919322967529297 ms per batch image class_id:0, confidence:0.86,left_top:[956.47,242.63], right_bottom:[1018.99,410.06] class_id:17, confidence:0.98,left_top:[0.75,348.76], right_bottom:[183.59,491.91] class_id:17, confidence:0.98,left_top:[838.18,281.39], right_bottom:[1128.14,484.16] class_id:17, confidence:0.98,left_top:[648.16,321.92], right_bottom:[853.18,479.21] class_id:17, confidence:0.85,left_top:[426.89,342.42], right_bottom:[638.06,498.02] class_id:17, confidence:0.79,left_top:[328.13,344.61], right_bottom:[469.42,482.50] detect frame:253 Inference: 20.92146873474121 ms per batch image class_id:0, confidence:0.90,left_top:[956.12,242.90], right_bottom:[1018.56,409.15] class_id:17, confidence:0.99,left_top:[0.00,346.85], right_bottom:[185.86,495.89] class_id:17, confidence:0.98,left_top:[835.83,278.02], right_bottom:[1127.18,488.58] class_id:17, confidence:0.98,left_top:[644.35,321.13], right_bottom:[855.50,480.76] class_id:17, confidence:0.94,left_top:[445.73,336.70], right_bottom:[650.15,486.24] class_id:17, confidence:0.81,left_top:[342.60,345.20], right_bottom:[479.72,481.89] detect frame:254 Inference: 20.951509475708008 ms per batch image class_id:0, confidence:0.95,left_top:[952.50,248.62], right_bottom:[1015.86,410.22] class_id:17, confidence:0.99,left_top:[837.66,278.20], right_bottom:[1122.63,489.95] class_id:17, confidence:0.98,left_top:[0.41,354.80], right_bottom:[184.95,501.62] class_id:17, confidence:0.98,left_top:[641.82,319.64], right_bottom:[853.41,482.97] class_id:17, confidence:0.89,left_top:[471.11,332.41], right_bottom:[644.34,489.31] class_id:17, confidence:0.70,left_top:[320.63,354.01], right_bottom:[470.06,487.61] detect frame:255 Inference: 20.973682403564453 ms per batch image class_id:0, confidence:0.95,left_top:[949.59,254.59], right_bottom:[1012.24,417.96] class_id:17, confidence:0.99,left_top:[822.88,283.57], right_bottom:[1114.85,489.52] class_id:17, confidence:0.98,left_top:[1.88,368.29], right_bottom:[175.82,512.56] class_id:17, confidence:0.98,left_top:[641.32,318.79], right_bottom:[851.31,485.68] class_id:17, confidence:0.94,left_top:[313.13,352.10], right_bottom:[475.74,495.47] class_id:17, confidence:0.93,left_top:[471.07,336.07], right_bottom:[640.00,491.21] class_id:17, confidence:0.26,left_top:[196.45,424.19], right_bottom:[264.05,458.48] class_id:17, confidence:0.25,left_top:[427.17,353.78], right_bottom:[543.44,492.73] detect frame:256 Inference: 21.055936813354492 ms per batch image class_id:0, confidence:0.94,left_top:[949.11,252.12], right_bottom:[1012.53,420.99] class_id:17, confidence:0.99,left_top:[3.00,369.19], right_bottom:[181.60,514.42] class_id:17, confidence:0.99,left_top:[837.38,278.03], right_bottom:[1106.29,491.40] class_id:17, confidence:0.97,left_top:[646.90,313.32], right_bottom:[853.69,483.24] class_id:17, confidence:0.92,left_top:[467.17,333.33], right_bottom:[640.19,491.04] class_id:17, confidence:0.91,left_top:[314.30,351.20], right_bottom:[475.05,493.08] detect frame:257 Inference: 20.919084548950195 ms per batch image class_id:0, confidence:0.88,left_top:[937.36,258.10], right_bottom:[1003.68,418.30] class_id:17, confidence:0.99,left_top:[1.83,370.86], right_bottom:[181.03,517.03] class_id:17, confidence:0.97,left_top:[829.47,281.21], right_bottom:[1104.19,493.77] class_id:17, confidence:0.91,left_top:[306.95,350.89], right_bottom:[479.85,493.36] class_id:17, confidence:0.91,left_top:[647.30,317.85], right_bottom:[851.46,485.20] class_id:17, confidence:0.79,left_top:[468.02,339.70], right_bottom:[635.95,499.94] detect frame:258 Inference: 21.00992202758789 ms per batch image class_id:0, confidence:0.84,left_top:[934.32,259.21], right_bottom:[998.58,416.03] class_id:17, confidence:0.98,left_top:[819.39,278.96], right_bottom:[1100.51,495.19] class_id:17, confidence:0.98,left_top:[3.60,374.91], right_bottom:[200.05,524.58] class_id:17, confidence:0.97,left_top:[301.52,356.07], right_bottom:[477.60,497.55] class_id:17, confidence:0.97,left_top:[625.73,321.12], right_bottom:[840.98,491.36] class_id:17, confidence:0.91,left_top:[464.80,343.54], right_bottom:[638.14,496.44] detect frame:259 Inference: 20.987272262573242 ms per batch image class_id:0, confidence:0.88,left_top:[930.99,264.04], right_bottom:[995.90,422.78] class_id:17, confidence:0.98,left_top:[814.30,284.60], right_bottom:[1101.03,494.57] class_id:17, confidence:0.98,left_top:[0.00,377.99], right_bottom:[183.90,523.17] class_id:17, confidence:0.96,left_top:[307.54,359.89], right_bottom:[479.22,500.99] class_id:17, confidence:0.92,left_top:[623.98,329.22], right_bottom:[838.69,491.16] class_id:17, confidence:0.92,left_top:[461.92,346.56], right_bottom:[637.63,508.04] detect frame:260 Inference: 20.824432373046875 ms per batch image class_id:0, confidence:0.92,left_top:[935.32,255.89], right_bottom:[998.49,417.59] class_id:17, confidence:0.99,left_top:[825.54,283.66], right_bottom:[1098.69,494.56] class_id:17, confidence:0.98,left_top:[0.00,368.30], right_bottom:[183.93,512.23] class_id:17, confidence:0.94,left_top:[313.35,353.67], right_bottom:[480.65,494.87] class_id:17, confidence:0.89,left_top:[627.47,326.44], right_bottom:[840.22,493.12] class_id:17, confidence:0.88,left_top:[423.89,347.37], right_bottom:[632.68,510.43] detect frame:261 Inference: 21.08931541442871 ms per batch image class_id:0, confidence:0.91,left_top:[932.99,266.10], right_bottom:[998.38,425.88] class_id:17, confidence:0.99,left_top:[821.74,293.49], right_bottom:[1102.11,500.98] class_id:17, confidence:0.96,left_top:[0.00,365.50], right_bottom:[189.56,510.38] class_id:17, confidence:0.95,left_top:[316.76,357.34], right_bottom:[479.83,496.75] class_id:17, confidence:0.92,left_top:[625.47,344.52], right_bottom:[835.42,502.07] class_id:17, confidence:0.84,left_top:[426.73,358.57], right_bottom:[633.17,509.86] detect frame:262 Inference: 21.12603187561035 ms per batch image class_id:0, confidence:0.87,left_top:[925.57,255.54], right_bottom:[986.52,431.84] class_id:17, confidence:0.99,left_top:[807.31,287.81], right_bottom:[1094.72,508.17] class_id:17, confidence:0.98,left_top:[0.00,365.00], right_bottom:[175.99,514.20] class_id:17, confidence:0.97,left_top:[625.52,336.11], right_bottom:[839.94,491.34] class_id:17, confidence:0.96,left_top:[302.65,354.18], right_bottom:[476.28,499.24] class_id:17, confidence:0.90,left_top:[427.20,356.85], right_bottom:[637.15,510.26] class_id:17, confidence:0.21,left_top:[474.90,350.07], right_bottom:[545.26,403.83] detect frame:263 Inference: 20.95627784729004 ms per batch image class_id:0, confidence:0.84,left_top:[925.80,260.21], right_bottom:[977.06,411.82] class_id:17, confidence:0.99,left_top:[1.20,372.98], right_bottom:[169.59,523.41] class_id:17, confidence:0.97,left_top:[792.68,289.44], right_bottom:[1078.02,507.84] class_id:17, confidence:0.92,left_top:[616.42,344.29], right_bottom:[816.49,503.47] class_id:17, confidence:0.86,left_top:[301.65,354.77], right_bottom:[466.94,509.13] class_id:17, confidence:0.79,left_top:[420.84,363.43], right_bottom:[631.18,516.19] class_id:17, confidence:0.33,left_top:[444.23,358.41], right_bottom:[539.27,408.75] detect frame:264 Inference: 20.977497100830078 ms per batch image class_id:0, confidence:0.72,left_top:[921.84,255.83], right_bottom:[972.72,414.37] class_id:17, confidence:0.99,left_top:[0.00,375.20], right_bottom:[175.02,527.28] class_id:17, confidence:0.98,left_top:[802.33,285.60], right_bottom:[1068.94,497.88] class_id:17, confidence:0.92,left_top:[611.44,337.98], right_bottom:[816.64,504.99] class_id:17, confidence:0.87,left_top:[462.96,352.56], right_bottom:[634.85,506.54] class_id:17, confidence:0.87,left_top:[297.23,352.61], right_bottom:[465.20,502.97] class_id:17, confidence:0.25,left_top:[422.91,351.48], right_bottom:[545.71,507.61] class_id:17, confidence:0.21,left_top:[439.72,358.80], right_bottom:[537.41,406.09] detect frame:265 Inference: 20.790815353393555 ms per batch image class_id:0, confidence:0.73,left_top:[904.11,260.69], right_bottom:[971.96,425.71] class_id:17, confidence:0.98,left_top:[795.51,284.67], right_bottom:[1065.81,501.45] class_id:17, confidence:0.98,left_top:[0.00,384.41], right_bottom:[168.32,530.06] class_id:17, confidence:0.93,left_top:[628.53,334.80], right_bottom:[810.31,490.97] class_id:17, confidence:0.90,left_top:[469.07,353.37], right_bottom:[627.93,506.38] class_id:17, confidence:0.82,left_top:[290.14,357.25], right_bottom:[466.62,506.07] detect frame:266 Inference: 20.826101303100586 ms per batch image class_id:0, confidence:0.68,left_top:[901.49,266.54], right_bottom:[968.21,426.54] class_id:17, confidence:0.98,left_top:[0.82,387.94], right_bottom:[166.25,532.08] class_id:17, confidence:0.95,left_top:[792.43,289.92], right_bottom:[1059.46,505.03] class_id:17, confidence:0.95,left_top:[608.30,339.48], right_bottom:[810.75,502.91] class_id:17, confidence:0.84,left_top:[293.53,363.73], right_bottom:[463.79,515.11] class_id:17, confidence:0.82,left_top:[416.86,359.65], right_bottom:[621.98,515.57] detect frame:267 Inference: 21.06165885925293 ms per batch image class_id:0, confidence:0.84,left_top:[895.70,262.51], right_bottom:[949.77,424.75] class_id:17, confidence:0.98,left_top:[0.00,389.47], right_bottom:[161.02,536.99] class_id:17, confidence:0.98,left_top:[773.91,288.64], right_bottom:[1044.34,502.03] class_id:17, confidence:0.95,left_top:[586.55,336.80], right_bottom:[798.45,505.50] class_id:17, confidence:0.92,left_top:[443.79,357.28], right_bottom:[614.35,505.03] class_id:17, confidence:0.81,left_top:[304.92,368.58], right_bottom:[457.51,513.59] detect frame:268 Inference: 21.164894104003906 ms per batch image class_id:0, confidence:0.85,left_top:[888.13,265.24], right_bottom:[943.93,409.57] class_id:17, confidence:0.99,left_top:[754.55,289.03], right_bottom:[1035.93,498.94] class_id:17, confidence:0.99,left_top:[0.00,392.87], right_bottom:[154.28,538.75] class_id:17, confidence:0.95,left_top:[441.42,355.55], right_bottom:[601.72,504.44] class_id:17, confidence:0.90,left_top:[595.42,328.72], right_bottom:[799.21,497.75] class_id:17, confidence:0.74,left_top:[283.24,372.29], right_bottom:[432.23,509.85] class_id:17, confidence:0.30,left_top:[396.31,371.55], right_bottom:[513.59,509.71]
from IPython.display import HTML
# bilibili
HTML('<iframe src="//player.bilibili.com/player.html?aid=753056627&bvid=BV1xk4y1k7j1&cid=187647930&page=1" scrolling="no" border="0" frameborder="no" framespacing="0" allowfullscreen="true"> </iframe>')<IPython.core.display.HTML object>
PaddleDetection实践应用
准备数据集
PaddleDetection的数据处理模块是一个Python模块，所有代码逻辑在ppdet/data/中，数据处理模块用于加载数据并将其转换成适用于物体检测模型的训练、评估、推理所需要的格式。 数据处理模块的主要构成如下架构所示。 PaddleDetection目前支持COCO、Pascal VOC 和WIDER-FACE数据源，默认数据集可自动下载，请参考默认数据集安装方法。同时还支持自定义数据源，包括(1)自定义数据源转换成COCO数据集；(2)定义新的数据源。详细操作请数据处理。
ppdet/data/├── reader.py # 数据处理模块的总接口├── shared_queue # 共享内存管理模块│ ├── queue.py # 定义共享内存队列│ ├── sharedmemory.py # 负责分配内存├── source # 数据源管理模块│ ├── dataset.py # 定义数据源基类，各类数据集继承于此│ ├── coco.py # COCO数据集解析与格式化数据│ ├── voc.py # Pascal VOC数据集解析与格式化数据│ ├── widerface.py # WIDER-FACE数据集解析与格式化数据├── tests # 单元测试模块│ ├── test_dataset.py # 对数据集解析、加载等进行单元测试│ │ ...├── tools # 一些有用的工具│ ├── x2coco.py # 将其他数据集转换为COCO数据集格式├── transform # 数据预处理模块│ ├── batch_operators.py # 定义各类基于批量数据的预处理算子│ ├── op_helper.py # 预处理算子的辅助函数│ ├── operators.py # 定义各类基于单张图片的预处理算子├── parallel_map.py # 在多进程/多线程模式中对数据预处理操作进行加速
模型训练
边训练边测试
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python -u tools/train.py -c configs/faster_rcnn_r50_1x.yml --eval 在训练中交替执行评估, 评估在每个snapshot_iter时开始。每次评估后还会评出最佳mAP模型保存到best_model文件夹下。
如果验证集很大，测试将会比较耗时，建议减少评估次数，或训练完再进行评估。
Fine-tune其他任务
使用预训练模型fine-tune其他任务时，可以直接加载预训练模型，形状不匹配的参数将自动忽略，例如：
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python -u tools/train.py -c configs/faster_rcnn_r50_1x.yml
-o pretrain_weights=output/faster_rcnn_r50_1x/model_final
也可以显示的指定忽略参数名，可采用如下两种方式：
在YAML配置文件中设置finetune_exclude_pretrained_params 在命令行中添加-o finetune_exclude_pretrained_params对预训练模型进行选择性加载。 export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python -u tools/train.py -c configs/faster_rcnn_r50_1x.yml
-o pretrain_weights=output/faster_rcnn_r50_1x/model_final
finetune_exclude_pretrained_params=[‘cls_score’,‘bbox_pred’] 详细说明请参考迁移学习文档
使用Paddle OP组建的YOLOv3损失函数训练YOLOv3。
为了便于用户重新设计修改YOLOv3的损失函数，我们也提供了不使用fluid.layer.yolov3_loss接口而是在python代码中使用Paddle OP的方式组建YOLOv3损失函数, 可通过如下命令用Paddle OP组建YOLOv3损失函数版本的YOLOv3模型：
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
python -u tools/train.py -c configs/yolov3_darknet.yml \-o use_fine_grained_loss=true
Paddle OP组建YOLOv3损失函数代码位于ppdet/modeling/losses/yolo_loss.py
CUDA_VISIBLE_DEVICES 参数可以指定不同的GPU。例如: export CUDA_VISIBLE_DEVICES=0,1,2,3.
若本地未找到数据集，将自动下载数据集并保存在~/.cache/paddle/dataset中。
预训练模型自动下载并保存在〜/.cache/paddle/weights中。
模型checkpoints默认保存在output中，可通过修改配置文件中save_dir进行配置。
混合精度训练
通过设置 --fp16 命令行选项可以启用混合精度训练。目前混合精度训练已经在Faster-FPN, Mask-FPN 及 Yolov3 上进行验证，几乎没有精度损失（小于0.2 mAP)。
建议使用多进程方式来进一步加速混合精度训练。示例如下。
python -m paddle.distributed.launch --selected_gpus 0,1,2,3,4,5,6,7 tools/train.py --fp16 -c configs/faster_rcnn_r50_fpn_1x.yml
如果训练过程中loss出现NaN，请尝试调节–loss_scale选项数值，细节请参看混合精度训练相关的Nvidia文档。
另外，请注意将配置文件中的 norm_type 由 affine_channel 改为 bn。
模型评估
指定权重和数据集路径
PaddleDetection目标检测任务简介和概述相关推荐
- PaddleDetection目标检测之水果检测（下）（yolov3_mobilenet_v1）
2021SC@SDUSC 接上篇博客 (3) JPEGImages:存储的就是实际的图片了 找一下apple_65.jpg看看 就是这个样子的 (4) create_list.py.label_lis ...
- 目标检测方法简介:RPN(Region Proposal Network) and SSD(Single Shot MultiBox Detector)
原文引用:http://lufo.me/2016/10/detection/ 最近几年深度学习在计算机视觉领域取得了巨大的成功,而在目标检测这一计算机视觉的经典问题上直到去年(2015)才有了完全使用 ...
- PaddleDetection目标检测之水果检测（上）（yolov3_mobilenet_v1）
2021SC@SDUSC 一.创建项目 (1)进入到https://aistudio.baidu.com/aistudio/projectoverview/public (2)创建项目 点击添加数据集 ...
- 2016CVPR目标检测论文简介
目标检测的指标: 1)识别精度 2)识别效率 3)定位准确性 CVPR2016专题: CVPR/ICCV目标检测最新论文 2016年的CVPR目标检测(这里讨论的是2D的目标检测)的方法主要是 基于C ...
- Udacity Self-Driving 目标检测数据集简介与使用
前言 之前,博主为了得到更好的车载视频目标检测效果(偏工程实际,非刷榜),使用SSD框架训练过KITTI数据集,几次训练下来,结果不太理想.自己分析,原因较多,其中很重要的一条就是KITTI数据集不够 ...
- 基于图像的目标检测与定位方法概述
目录 1. 目标检测与定位概念 2. 目标检测与定位方法 2.1 传统目标检测流程 2.2 two-stage检测算法 2.2.1 R-CNN 2.2.2 two-stage其他算法 2.2.2.1 ...
- CV之OD：计算机视觉之目标检测(Object Detection)方向的简介、使用方法、案例应用之详细攻略
CV之OD:计算机视觉之目标检测(Object Detection)方向的简介.使用方法.案例应用之详细攻略 目录 OD目标检测的简介 1.Selective Search for Object Re ...
- 动态视频目标检测和跟踪技术(入门)
动态视频目标检测和跟踪技术 http://m.qingqingsk.com/ztnews/lvvozlzrztkzrqwqqlnrluqk.html 传统电视监控技术只能达到"千里眼&quo ...
- 如何攻克目标检测？百度全球顶级会议双料冠军团队有话说！
计算机视觉最火方向是什么? 当然是目标检测啦! 目标检测,是计算机视觉领域的核心问题之一,近两年全球顶会的相关论文达上百篇,受到越来越多的人关注.无论是做人脸识别.自动驾驶.文字检测.人机交互,都离不 ...
最新文章
- 线索二叉树原理及前序、中序线索化（Java版）
- python调用c代码
- 男人是消耗品，女人是战利品？
- 楚乔传手游 服务器维护,楚乔传手游网络异常进不去游戏怎么办？楚乔传网络异常解决方法...
- 从资源管理器中，获取被选择的文件的路径（及文件夹）的API
- lisp 多段线转面域_Objectarx 相交矩形求并集 面域转多段线
- jQuery JCrop插件的一个问题
- poj 3694 Network (无向图的 割边 lca )
- ironpython使用dictionary_Ironpython引用C#数据结构
- 免费而强大的十款PCB设计软件
- ubuntu系统firefox浏览器无法播放音视频
- 靶机渗透练习21-Noob
- 栈的初识（C语言）（1）
- 企业必须面对云计算时代
- 数学模型参考文献的格式
- [转] 关于“时间”的一次探索
- Eclipse 插件大全
- JS获取时间戳 10位数时间戳
- 10个建模师，9个吃不饱，游戏建模师职业现状
- 小侃设计模式（十五）-命令模式