[OpenMMLab]AI实战营第七节课

语义分割代码实战教学

HRNet 高分辨率神经网络

安装配置

# 选择分支
git branch -a
git switch 3.x
# 配置环境
conda create -n mmsegmentation python=3.8
conda activate mmsegmentation
pip install torch==1.11.0+cu113 torchvision==0.12.0+cu113 torchaudio==0.11.0 --extra-index-url https://download.pytorch.org/whl/cu113
pip install mmcv==2.0.0rc3 -f https://download.openmmlab.com/mmcv/dist/cu113/torch1.11/index.html
pip install -U openmim
mim install mmengine
pip install -v -e .
# 下载预训练模型
wget https://download.openmmlab.com/mmsegmentation/v0.5/pspnet/pspnet_r50-d8_512x1024_40k_cityscapes/pspnet_r50-d8_512x1024_40k_cityscapes_20200605_003338-2966598c.pth

预训练语义分割模型预测图片

通过脚本文件，利用预训练模型进行预测

python demo/image_demo.py \data/street_uk.jpeg \configs/pspnet/pspnet_r50-d8_4xb2-40k_cityscapes-512x1024.py \https://download.openmmlab.com/mmsegmentation/v0.5/pspnet/pspnet_r50-d8_512x1024_40k_cityscapes/pspnet_r50-d8_512x1024_40k_cityscapes_20200605_003338-2966598c.pth \--out-file outputs/B1_uk_pspnet.jpg \--device cuda:0 \--opacity 0.5

--opacity的作用是调节透明度，更像原图或者更像语义分割后的图

通过编写api来提取信息

from mmseg.apis import init_model
model = init_model(config_file, checkpoint_file, device='cuda:0')
result = inference_model(model, img_path)
result

result.keys()
>>> ['pred_sem_seg', 'seg_logits']# result.pred_sem_seg中语义分割图为单通道图，每个值为0-18，即共19各类别
result.pred_sem_seg.data.shape
>>> torch.Size([1, 1500, 2250])# 一共多少类别
np.unique(result.pred_sem_seg.data.cpu())
>>> array([ 0,  1,  2,  3,  4,  5,  6,  7,  8, 10, 11, 13, 15])# result.seg_logits是置信度，每一个像素属于预测类别的置信度
result.seg_logits.data.shape
>>> torch.Size([19, 1500, 2250])

预训练语义分割模型预测视频

python demo/video_demo.py \data/traffic.mp4 \configs/pspnet/pspnet_r50-d8_4xb2-40k_cityscapes-512x1024.py \https://download.openmmlab.com/mmsegmentation/v0.5/pspnet/pspnet_r50-d8_512x1024_40k_cityscapes/pspnet_r50-d8_512x1024_40k_cityscapes_20200605_003338-2966598c.pth \--device cuda:0 \--output-file outputs/B3_video.mp4 \--opacity 0.5

在自己数据集上训练语义分割模型

下载数据集

wget https://zihao-openmmlab.obs.cn-east-3.myhuaweicloud.com/20230130-mmseg/dataset/iccv09Data.tar.gz -O stanford_background.tar.gz

修改数据集类

from mmseg.registry import DATASETS
from mmseg.datasets import [e-mail]_module()
class StanfordBackgroundDataset(BaseSegDataset):METAINFO = dict(classes = classes, palette = palette)def __init__(self, **kwargs):super().__init__(img_suffix='.jpg', seg_map_suffix='.png', **kwargs)

修改config配置文件

1. 修改model.head.num_classes
2. 修改数据集的data_type和data_root
3. 指定训练集的路径和测试集的路径
4. 指定预训练模型权重文件路径
5. 修改训练配置参数，训练epoch，batch_size等