霹雳吧啦Wz语义分割学习笔记P10

P10.LR-ASPP源码讲解

train predict与fcn源码章节相同，唯一不同的是create_model创建的是lraspp结构（在src文件夹中的lraspp_model）

def lraspp_mobilenetv3_large(num_classes=21, pretrain_backbone=False):#定义函数实例化lraspp模型# 'mobilenetv3_large_imagenet': 'https://download.pytorch.org/models/mobilenet_v3_large-8738ca79.pth'# 'lraspp_mobilenet_v3_large_coco': 'https://download.pytorch.org/models/lraspp_mobilenet_v3_large-d234d4ea.pth'backbone = mobilenet_v3_large(dilated=True)#需要使用膨胀卷积if pretrain_backbone:# 载入mobilenetv3 large backbone预训练权重backbone.load_state_dict(torch.load("mobilenet_v3_large.pth", map_location='cpu'))backbone = backbone.features# Gather the indices of blocks which are strided. These are the locations of C1, ..., Cn-1 blocks.# The first and last blocks are always included because they are the C0 (conv1) and Cn.stage_indices = [0] + [i for i, b in enumerate(backbone) if getattr(b, "is_strided", False)] + [len(backbone) - 1]low_pos = stage_indices[-4]  # use C2 here which has output_stride = 8high_pos = stage_indices[-1]  # use C5 which has output_stride = 16low_channels = backbone[low_pos].out_channelshigh_channels = backbone[high_pos].out_channelsreturn_layers = {str(low_pos): "low", str(high_pos): "high"}backbone = IntermediateLayerGetter(backbone, return_layers=return_layers)model = LRASPP(backbone, low_channels, high_channels, num_classes)return model

霹雳吧啦Wz语义分割学习笔记P10相关推荐

霹雳吧啦Wz语义分割学习笔记P3
P3.FCN网络结构详解 1.前言 Fully Convolutional Networks for Semantic Segmentation 2015CVPR https://arxiv.org/ ...
霹雳吧啦Wz语义分割学习笔记P6
P6.DeepLabV1网络简介 1.前言 semantic image segmentation with deep convolutional nets and fully connected C ...
霹雳吧啦Wz语义分割学习笔记P2
P2.转置卷积Transposed Convolution 1.作用:上采样转置卷积不是卷积的逆运算转置卷积也是卷积 2.转置卷积运算步骤在输入特征图元素间填充s-1行.列0 在输入特征图四周填 ...
霹雳吧啦Wz语义分割学习笔记P8
P8.DeepLabV3网络简介 1.前言 Rethinking Astrous Convolution for Semantic Image Segmentation 2017CVPR https: ...
霹雳吧啦Wz语义分割学习笔记P5
P5.Pytorch官方实现的FCN网络结构 backbone:VGG16 -> ResNet50:使用了膨胀卷积
使用Pytorch搭建U-Net网络并基于DRIVE数据集训练(语义分割)学习笔记
使用Pytorch搭建U-Net网络并基于DRIVE数据集训练(语义分割)学习笔记 https://www.bilibili.com/video/BV1rq4y1w7xM?spm_id_from=33 ...
语义分割学习笔记（四）膨胀卷积
推荐课程: 1.膨胀卷积_哔哩哔哩_bilibili(膨胀卷积原理讲的很清楚) 2.膨胀卷积(Dilated convolution)详解_哔哩哔哩_bilibili(gradding effect问 ...
联合目标检测和语义分割——学习笔记
联合目标检测和语义分割目标检测目标检测是一种与计算机视觉和图像处理相关的计算机技术,用于检测数字图像和视频中特定类别的语义对象(例如人,建筑物或汽车)的实例.然而现实中物体的尺寸.姿态.位置都有很 ...
语义分割学习笔记（一）
文章目录一.区别二.代码演示原图像与mask融合三.数据处理一.区别语义分割:每个像素都打上标签(这个像素是人,树,背景等)语义分割只区分类别,不区分类别中的具体单元实例分割:不光要区分类 ...
点云分割学习笔记2022
PointCloudSegmentation 点云分割 PointCloudSegmentation测试笔记_AI视觉网奇的博客-CSDN博客代码地址:https://github.com/Liya ...

霹雳吧啦Wz语义分割学习笔记P10

P10.LR-ASPP源码讲解

霹雳吧啦Wz语义分割学习笔记P10相关推荐

最新文章

热门文章