安装使用了一波detectron2 ，感觉没有mmdetection好用…但是还是记一下安装和使用的过程

安装

确保你的python版本是3.6以上

查看你的机器的cuda版本(如果没安装的话安装)，通过

nvcc --version

命令查看，然后去安装或者更新相应版本的pytorch。

比如我的cuda版本是10.0的，就执行以下命令安装或者更新对应cuda版本的pytorch，不然会报错

pip3 install torch==1.4.0+cu100 torchvision==0.5.0+cu100 -f https://download.pytorch.org/whl/torch_stable.html

然后继续安装

pip3 install opencv-python

pip3 install cython pyyaml==5.1

pip3 install -U 'git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI'

然后clone

git clone https://github.com/zhanghang1989/detectron2-ResNeSt.git

编译

cd detectron2 && python3 -m pip install -e .

成功！

运行

# You may need to restart your runtime prior to this, to let your installation take effect

# Some basic setup:

# Setup detectron2 logger

import detectron2

from detectron2.utils.logger import setup_logger

setup_logger()

# import some common libraries

import numpy as np

import cv2

import os

import random

from matplotlib import pyplot as plt

# import some common detectron2 utilities

from detectron2 import model_zoo

from detectron2.engine import DefaultPredictor

from detectron2.config import get_cfg

from detectron2.utils.visualizer import Visualizer

from detectron2.data import MetadataCatalog

from detectron2.data import DatasetCatalog

#

from detectron2.data.datasets import register_coco_instances

先要注册你的COCO格式的数据集，比如我有一份训练集train.json和验证集val.json，我在detectron2里注册，并分别命名为mobile_train和mobile_val。

register_coco_instances("mobile_train", {}, "train.json", "/workspace/nas-data/train/")

register_coco_instances("mobile_val", {}, "/workspace/nas-data/fuck_china_moble/mnt/3/fuck_chinMobile_val.json", "/workspace/nas-data/train/")

说明一下

"/workspace/nas-data/train/"

是我的图像所在的绝对路径，比如我的annotations里面的images里面的是这样的

{

"file_name": "train_r1062.jpeg",

"height": 216,

"width": 235,

"id": 1048

},

那么

/workspace/nas-data/train/train_r1062.jpeg

这个就是我的这个图像的路径

可以看一下你的数据集有没有正确注册，通过下面这个代码检查一下

dataset_dicts = DatasetCatalog.get("mobile_val")

balloon_metadata = MetadataCatalog.get("mobile_val")

for d in random.sample(dataset_dicts, 1):

img = cv2.imread(d["file_name"])

visualizer = Visualizer(img[:, :, ::-1], metadata=balloon_metadata, scale=0.5)

vis = visualizer.draw_dataset_dict(d)

plt.figure(figsize=(12,12))

plt.imshow(vis.get_image()[:, :, ::-1])

就会输出

ok我们开始训练

from detectron2.engine import DefaultTrainer

from detectron2.config import get_cfg

import os

cfg = get_cfg()

#配置文件路径

cfg.merge_from_file(

"/root/detectron2-ResNeSt/configs/COCO-Detection/faster_cascade_rcnn_ResNeSt_200_FPN_syncbn_range-scale_1x.yaml"

)

cfg.DATASETS.TRAIN = ("mobile_train",)

cfg.DATASETS.TEST = ()

cfg.DATALOADER.NUM_WORKERS = 2

#预训练文件

cfg.MODEL.WEIGHTS = "/root/detectron2-ResNeSt/faster_cascade_rcnn_ResNeSt_200_FPN_syncbn_range-scale_1x-1be2a87e.pth" # initialize from model zoo

cfg.SOLVER.IMS_PER_BATCH = 4

cfg.SOLVER.BASE_LR = 0.00125 * 1 * 4

cfg.SOLVER.MAX_ITER = (int(34*8*89*2/4))

cfg.SOLVER.STEPS = (int(24*8*89*2/4), int(33*8*89*2/4))

cfg.MODEL.ROI_HEADS.BATCH_SIZE_PER_IMAGE = (

512

) # faster, and good enough for this toy dataset

cfg.MODEL.ROI_HEADS.NUM_CLASSES = 3

#输出权重目录

cfg.OUTPUT_DIR = "/root/detectron2-ResNeSt/output/baseline"

os.makedirs(cfg.OUTPUT_DIR, exist_ok=True)

trainer = DefaultTrainer(cfg)

trainer.resume_or_load(resume=False)

trainer.train()

训练完测试

cfg.MODEL.WEIGHTS = os.path.join(cfg.OUTPUT_DIR, "model_final.pth")

cfg.MODEL.ROI_HEADS.SCORE_THRESH_TEST = 0.5 # set the testing threshold for this model

cfg.DATASETS.TEST = ("mobile_val", )

predictor = DefaultPredictor(cfg)

from detectron2.utils.visualizer import ColorMode

for d in random.sample(dataset_dicts, 1):

im = cv2.imread(d["file_name"])

outputs = predictor(im)

v = Visualizer(im[:, :, ::-1],

metadata=balloon_metadata,

scale=0.8,

# instance_mode=ColorMode.IMAGE_BW # remove the colors of unsegmented pixels

)

v = v.draw_instance_predictions(outputs["instances"].to("cpu"))

plt.imshow(v.get_image()[:, :, ::-1])

看mAP

from detectron2.evaluation import COCOEvaluator, inference_on_dataset

from detectron2.data import build_detection_test_loader

evaluator = COCOEvaluator("mobile_val", cfg, False, output_dir="./output/")

val_loader = build_detection_test_loader(cfg, "mobile_val")

inference_on_dataset(trainer.model, val_loader, evaluator)

DONE (t=0.05s).

Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.334

Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.655

Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.285

Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.305

Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.469

Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.471

Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.250

Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.462

Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.463

Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.412

Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.551

Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.490