AI实战营第二期第八节《MMSegmentation代码课》

【课程链接】https://www.bilibili.com/video/BV1uh411T73q/
【讲师介绍】张子豪 OpenMMLab算法工程师
【学习形式】录播+社群答疑
【作业布置】本次课程为实战课，需提交笔记+作业。
在这里插入图片描述

课程大纲：

环境配置
预训练模型预测图片、视频
航拍图像语义分割案例
肾小球病理切片语义分割案例

作业：

西瓜瓤、西瓜皮、西瓜籽像素级语义分割

安装配置MMSegmentation

安装pytorch

pip install install torch==1.10.1+cu113 torchvision==0.11.2+cu113 torchaudio==0.10.1+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html

用MIM安装MMCV

pip install -U openmim
mim install mmengine
mim install 'mmcv==2.0.0rc4'

安装其它工具包

pip install opencv-python pillow matplotlib seaborn tqdm pytorch-lightning 'mmdet>=3.0.0rc1' -i https://pypi.tuna.tsinghua.edu.cn/simple

下载与安装MMSegmentation

# 删掉原有的 mmsegmentation 文件夹（如有）
rm -rf mmsegmentation
# 从 github 上下载最新的 mmsegmentation 源代码
git clone https://github.com/open-mmlab/mmsegmentation.git -b dev-1.x
pip install -v -e .

下载预训练模型权重文件和视频素材

import os

# 创建 checkpoint 文件夹，用于存放预训练模型权重文件
os.mkdir('checkpoint')

# 创建 outputs 文件夹，用于存放预测结果
os.mkdir('outputs')

# 创建 data 文件夹，用于存放图片和视频素材
os.mkdir('data')

下载预训练模型权重至checkpoint目录

Model Zoo：https://github.com/open-mmlab/mmsegmentation/blob/master/docs/en/model_zoo.md

 从 Model Zoo 获取 PSPNet 预训练模型，下载并保存在 checkpoint 文件夹中
!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 -P checkpoint

下载素材至data目录

如果报错Unable to establish SSL connection.，重新运行代码块即可。

# 伦敦街景图片
!wget https://zihao-openmmlab.obs.cn-east-3.myhuaweicloud.com/20220713-mmdetection/images/street_uk.jpeg -P data

# 上海驾车街景视频，视频来源：https://www.youtube.com/watch?v=ll8TgCZ0plk
!wget https://zihao-download.obs.cn-east-3.myhuaweicloud.com/detectron2/traffic.mp4 -P data

# 街拍视频，2022年3月30日
!wget https://zihao-openmmlab.obs.cn-east-3.myhuaweicloud.com/20220713-mmdetection/images/street_20220330_174028.mp4 -P data

检查安装成功

# 检查 Pytorch
import torch, torchvision
print('Pytorch 版本', torch.__version__)
print('CUDA 是否可用',torch.cuda.is_available())

# 检查 mmcv
import mmcv
from mmcv.ops import get_compiling_cuda_version, get_compiler_version
print('MMCV版本', mmcv.__version__)
print('CUDA版本', get_compiling_cuda_version())
print('编译器版本', get_compiler_version())

# 检查 mmsegmentation
import mmseg
from mmseg.utils import register_all_modules
from mmseg.apis import inference_model, init_model
print('mmsegmentation版本', mmseg.__version__)

设置Matplotlib中文字体

# # windows操作系统
# plt.rcParams['font.sans-serif']=['SimHei']  # 用来正常显示中文标签 
# plt.rcParams['axes.unicode_minus']=False  # 用来正常显示负号
# Mac操作系统，参考 https://www.ngui.cc/51cto/show-727683.html
# 下载 simhei.ttf 字体文件
# !wget https://zihao-openmmlab.obs.cn-east-3.myhuaweicloud.com/20220716-mmclassification/dataset/SimHei.ttf

# Linux操作系统，例如 云GPU平台：https://featurize.cn/?s=d7ce99f842414bfcaea5662a97581bd1
# 如果遇到 SSL 相关报错，重新运行本代码块即可
!wget https://zihao-openmmlab.obs.cn-east-3.myhuaweicloud.com/20220716-mmclassification/dataset/SimHei.ttf -O /environment/miniconda3/lib/python3.7/site-packages/matplotlib/mpl-data/fonts/ttf/SimHei.ttf
!rm -rf /home/featurize/.cache/matplotlib

import matplotlib 
import matplotlib.pyplot as plt
matplotlib.rc("font",family='SimHei') # 中文字体

plt.plot([1,2,3], [100,500,300])
plt.title('matplotlib中文字体测试', fontsize=25)
plt.xlabel('X轴', fontsize=15)
plt.ylabel('Y轴', fontsize=15)
plt.show()

预训练语义分割模型预测

进入 mmsegmentation 主目录

import os
os.chdir('mmsegmentation')

载入测试图像

from PIL import Image

MMSegmentation模型库

Model Zoo：https://github.com/open-mmlab/mmsegmentation/blob/master/docs/en/model_zoo.md

常用config和checkpoint文件

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

configs/segformer/segformer_mit-b5_8xb1-160k_cityscapes-1024x1024.py

https://download.openmmlab.com/mmsegmentation/v0.5/segformer/segformer_mit-b5_8x1_1024x1024_160k_cityscapes/segformer_mit-b5_8x1_1024x1024_160k_cityscapes_20211206_072934-87a052ec.pth

configs/mask2former/mask2former_swin-l-in22k-384x384-pre_8xb2-90k_cityscapes-512x1024.py

https://download.openmmlab.com/mmsegmentation/v0.5/mask2former/mask2former_swin-l-in22k-384x384-pre_8xb2-90k_cityscapes-512x1024/mask2former_swin-l-in22k-384x384-pre_8xb2-90k_cityscapes-512x1024_20221202_141901-28ad20f1.pth

PSPNet

PSPNet语义分割算法

!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

在这里插入图片描述

SegFormer

!python demo/image_demo.py \
        data/street_uk.jpeg \
        configs/segformer/segformer_mit-b5_8xb1-160k_cityscapes-1024x1024.py \
        https://download.openmmlab.com/mmsegmentation/v0.5/segformer/segformer_mit-b5_8x1_1024x1024_160k_cityscapes/segformer_mit-b5_8x1_1024x1024_160k_cityscapes_20211206_072934-87a052ec.pth \
        --out-file outputs/B1_uk_segformer.jpg \
        --device cuda:0 \
        --opacity 0.5

在这里插入图片描述

Mask2Former

!python demo/image_demo.py \
        data/street_uk.jpeg \
        configs/mask2former/mask2former_swin-l-in22k-384x384-pre_8xb2-90k_cityscapes-512x1024.py \
        https://download.openmmlab.com/mmsegmentation/v0.5/mask2former/mask2former_swin-l-in22k-384x384-pre_8xb2-90k_cityscapes-512x1024/mask2former_swin-l-in22k-384x384-pre_8xb2-90k_cityscapes-512x1024_20221202_141901-28ad20f1.pth \
        --out-file outputs/B1_uk_Mask2Former.jpg \
        --device cuda:0 \
        --opacity 0.5

ADE20K语义分割数据集

mmsegmentation/mmseg/datasets/ade.py

关于ADE20K的故事：https://www.zhihu.com/question/390783647/answer/1226097849

!python demo/image_demo.py \
        data/street_uk.jpeg \
        configs/segformer/segformer_mit-b5_8xb2-160k_ade20k-640x640.py \
        https://download.openmmlab.com/mmsegmentation/v0.5/segformer/segformer_mit-b5_640x640_160k_ade20k/segformer_mit-b5_640x640_160k_ade20k_20220617_203542-940a6bd8.pth \
        --out-file outputs/B1_Segformer_ade20k.jpg \
        --device cuda:0 \
        --opacity 0.5

在这里插入图片描述

预训练语义分割模型预测-视频

视频预测-命令行（不推荐，慢）

!python demo/video_demo.py \
        data/street_20220330_174028.mp4 \
        configs/mask2former/mask2former_swin-l-in22k-384x384-pre_8xb2-90k_cityscapes-512x1024.py \
        https://download.openmmlab.com/mmsegmentation/v0.5/mask2former/mask2former_swin-l-in22k-384x384-pre_8xb2-90k_cityscapes-512x1024/mask2former_swin-l-in22k-384x384-pre_8xb2-90k_cityscapes-512x1024_20221202_141901-28ad20f1.pth \
        --device cuda:0 \
        --output-file outputs/B3_video.mp4 \
        --opacity 0.5

视频预测-Python API（推荐，快）

import numpy as np
import time
import shutil

import torch

from PIL import Image
import cv2

import mmcv
import mmengine
from mmseg.apis import inference_model
from mmseg.utils import register_all_modules
register_all_modules()

from mmseg.datasets import CityscapesDataset

# 模型 config 配置文件
config_file = 'configs/mask2former/mask2former_swin-l-in22k-384x384-pre_8xb2-90k_cityscapes-512x1024.py'

# 模型 checkpoint 权重文件
checkpoint_file = 'https://download.openmmlab.com/mmsegmentation/v0.5/mask2former/mask2former_swin-l-in22k-384x384-pre_8xb2-90k_cityscapes-512x1024/mask2former_swin-l-in22k-384x384-pre_8xb2-90k_cityscapes-512x1024_20221202_141901-28ad20f1.pth'
from mmseg.apis import init_model
model = init_model(config_file, checkpoint_file, device='cuda:0')

from mmengine.model.utils import revert_sync_batchnorm
if not torch.cuda.is_available():
    model = revert_sync_batchnorm(model)
# input_video = 'data/traffic.mp4'

input_video = 'data/street_20220330_174028.mp4'

temp_out_dir = time.strftime('%Y%m%d%H%M%S')
os.mkdir(temp_out_dir)
print('创建临时文件夹 {} 用于存放每帧预测结果'.format(temp_out_dir))
# 获取 Cityscapes 街景数据集 类别名和调色板
from mmseg.datasets import cityscapes
classes = cityscapes.CityscapesDataset.METAINFO['classes']
palette = cityscapes.CityscapesDataset.METAINFO['palette']
def pridict_single_frame(img, opacity=0.2):
    
    result = inference_model(model, img)
    
    # 将分割图按调色板染色
    seg_map = np.array(result.pred_sem_seg.data[0].detach().cpu().numpy()).astype('uint8')
    seg_img = Image.fromarray(seg_map).convert('P')
    seg_img.putpalette(np.array(palette, dtype=np.uint8))
    
    show_img = (np.array(seg_img.convert('RGB')))*(1-opacity) + img*opacity
    
    return show_img
# 读入待预测视频
imgs = mmcv.VideoReader(input_video)

prog_bar = mmengine.ProgressBar(len(imgs))

# 对视频逐帧处理
for frame_id, img in enumerate(imgs):
    
    ## 处理单帧画面
    show_img = pridict_single_frame(img, opacity=0.15)
    temp_path = f'{temp_out_dir}/{frame_id:06d}.jpg' # 保存语义分割预测结果图像至临时文件夹
    cv2.imwrite(temp_path, show_img)

    prog_bar.update() # 更新进度条

# 把每一帧串成视频文件
mmcv.frames2video(temp_out_dir, 'outputs/B3_video.mp4', fps=imgs.fps, fourcc='mp4v')

shutil.rmtree(temp_out_dir) # 删除存放每帧画面的临时文件夹
print('删除临时文件夹', temp_out_dir)

Kaggle实战-迪拜卫星航拍多类别语义分割

下载整理好的数据集

Kaggle原版数据集：https://www.kaggle.com/datasets/humansintheloop/semantic-segmentation-of-aerial-imagery

下载整理好之后的数据集

!rm -rf Dubai-dataset.zip Dubai-dataset

!wget https://zihao-openmmlab.obs.cn-east-3.myhuaweicloud.com/20230130-mmseg/dataset/Dubai-dataset.zip

!unzip Dubai-dataset.zip >> /dev/null # 解压

!rm -rf Dubai-dataset.zip # 删除压缩包

删除系统自动生成的多余文件

查看待删除的多余文件

!find . -iname '__MACOSX'
!find . -iname '__MACOSX'
!find . -iname '.DS_Store'
!find . -iname '.ipynb_checkpoints'
./.ipynb_checkpoints

删除多余文件

!for i in `find . -iname '__MACOSX'`; do rm -rf $i;done
!for i in `find . -iname '.DS_Store'`; do rm -rf $i;done
!for i in `find . -iname '.ipynb_checkpoints'`; do rm -rf $i;done

验证多余文件已删除

!find . -iname '__MACOSX'
!find . -iname '.DS_Store'
!find . -iname '.ipynb_checkpoints'

可视化探索数据集

import os

import cv2
import numpy as np
from PIL import Image
from tqdm import tqdm

import matplotlib.pyplot as plt
%matplotlib inline

# 指定单张图像路径
img_path = 'Dubai-dataset/img_dir/train/14.jpg'
mask_path = 'Dubai-dataset/ann_dir/train/14.png'

img = cv2.imread(img_path)
mask = cv2.imread(mask_path)
# 可视化语义分割标注
plt.imshow(mask[:,:,0])
plt.show()

在这里插入图片描述

plt.imshow(img[:,:,::-1])
plt.imshow(mask[:,:,0], alpha=0.4) # alpha 高亮区域透明度，越小越接近原图
plt.axis('off')
plt.show()

在这里插入图片描述
批量可视化图像和标注

# 指定图像和标注路径
PATH_IMAGE = 'Dubai-dataset/img_dir/train'
PATH_MASKS = 'Dubai-dataset/ann_dir/train'
# n行n列可视化
n = 5

# 标注区域透明度
opacity = 0.5

fig, axes = plt.subplots(nrows=n, ncols=n, sharex=True, figsize=(12,12))

for i, file_name in enumerate(os.listdir(PATH_IMAGE)[:n**2]):
    
    # 载入图像和标注
    img_path = os.path.join(PATH_IMAGE, file_name)
    mask_path = os.path.join(PATH_MASKS, file_name.split('.')[0]+'.png')
    img = cv2.imread(img_path)
    mask = cv2.imread(mask_path)
    
    # 可视化
    axes[i//n, i%n].imshow(img)
    axes[i//n, i%n].imshow(mask[:,:,0], alpha=opacity)
    axes[i//n, i%n].axis('off') # 关闭坐标轴显示
fig.suptitle('Image and Semantic Label', fontsize=30)
plt.tight_layout()
plt.show()

准备config配置文件

import numpy as np
from PIL import Image

import os.path as osp
from tqdm import tqdm

import mmcv
import mmengine
import matplotlib.pyplot as plt
%matplotlib inline

定义数据集类（各类别名称及配色）

!rm -rf mmseg/datasets/DubaiDataset.py # 删除原有文件
!wget https://zihao-openmmlab.obs.cn-east-3.myhuaweicloud.com/20230130-mmseg/Dubai/DubaiDataset.py -P mmseg/datasets

注册数据集类

!rm -rf mmseg/datasets/__init__.py # 删除原有文件
!wget https://zihao-openmmlab.obs.cn-east-3.myhuaweicloud.com/20230130-mmseg/Dubai/__init__.py -P mmseg/datasets

定义训练及测试pipeline

!rm -rf configs/_base_/datasets/DubaiDataset_pipeline.py
!wget https://zihao-openmmlab.obs.cn-east-3.myhuaweicloud.com/20230130-mmseg/Dubai/DubaiDataset_pipeline.py -P configs/_base_/datasets

下载模型config配置文件

!rm -rf configs/pspnet/pspnet_r50-d8_4xb2-40k_DubaiDataset.py # 删除原有文件
!wget https://zihao-openmmlab.obs.cn-east-3.myhuaweicloud.com/20230130-mmseg/Dubai/pspnet_r50-d8_4xb2-40k_DubaiDataset.py -P configs/pspnet

载入config配置文件

from mmengine import Config
cfg = Config.fromfile('./configs/pspnet/pspnet_r50-d8_4xb2-40k_DubaiDataset.py')

修改config配置文件

cfg.norm_cfg = dict(type='BN', requires_grad=True) # 只使用GPU时，BN取代SyncBN
cfg.crop_size = (256, 256)
cfg.model.data_preprocessor.size = cfg.crop_size
cfg.model.backbone.norm_cfg = cfg.norm_cfg
cfg.model.decode_head.norm_cfg = cfg.norm_cfg
cfg.model.auxiliary_head.norm_cfg = cfg.norm_cfg
# modify num classes of the model in decode/auxiliary head

# 模型 decode/auxiliary 输出头，指定为类别个数
cfg.model.decode_head.num_classes = 6
cfg.model.auxiliary_head.num_classes = 6

cfg.train_dataloader.batch_size = 8

cfg.test_dataloader = cfg.val_dataloader

# 结果保存目录
cfg.work_dir = './work_dirs/DubaiDataset'

# 训练迭代次数
cfg.train_cfg.max_iters = 3000
# 评估模型间隔
cfg.train_cfg.val_interval = 400
# 日志记录间隔
cfg.default_hooks.logger.interval = 100
# 模型权重保存间隔
cfg.default_hooks.checkpoint.interval = 1500

# 随机数种子
cfg['randomness'] = dict(seed=0)

查看完整config配置文件

print(cfg.pretty_text)

保存config配置文件

cfg.dump('pspnet-DubaiDataset_20230612.py')

MMSegmentation训练语义分割模型

import numpy as np

import os.path as osp
from tqdm import tqdm

import mmcv
import mmengine

from mmengine import Config
cfg = Config.fromfile('pspnet-DubaiDataset_20230612.py')

from mmengine.runner import Runner
from mmseg.utils import register_all_modules

# register all modules in mmseg into the registries
# do not init the default scope here because it will be init in the runner
register_all_modules(init_default_scope=False)
runner = Runner.from_cfg(cfg)

runner.train()

开始训练

如果遇到报错CUDA out of memeory，可尝试以下步骤：

调小 batch size
左上角内核-关闭所有内核
重启实例，或者使用显存更高的实例即可。

可视化训练日志

设置Matplotlib中文字体

# # windows操作系统
# plt.rcParams['font.sans-serif']=['SimHei']  # 用来正常显示中文标签 
# plt.rcParams['axes.unicode_minus']=False  # 用来正常显示负号
# Mac操作系统，参考 https://www.ngui.cc/51cto/show-727683.html
# 下载 simhei.ttf 字体文件
# !wget https://zihao-openmmlab.obs.cn-east-3.myhuaweicloud.com/20220716-mmclassification/dataset/SimHei.ttf

# Linux操作系统，例如 云GPU平台：https://featurize.cn/?s=d7ce99f842414bfcaea5662a97581bd1
# 如果遇到 SSL 相关报错，重新运行本代码块即可
!wget https://zihao-openmmlab.obs.cn-east-3.myhuaweicloud.com/20220716-mmclassification/dataset/SimHei.ttf -O /environment/miniconda3/lib/python3.7/site-packages/matplotlib/mpl-data/fonts/ttf/SimHei.ttf
!rm -rf /home/featurize/.cache/matplotlib

import matplotlib 
import matplotlib.pyplot as plt
matplotlib.rc("font",family='SimHei') # 中文字体

plt.plot([1,2,3], [100,500,300])
plt.title('matplotlib中文字体测试', fontsize=25)
plt.xlabel('X轴', fontsize=15)
plt.ylabel('Y轴', fontsize=15)
plt.show()

导入工具包
import pandas as pd

import matplotlib.pyplot as plt
%matplotlib inline
载入训练日志

日志文件路径

log_path = 'work_dirs/DubaiDataset/20230612_100725/vis_data/scalars.json'
with open(log_path, "r") as f:
    json_list = f.readlines()
len(json_list)

eval(json_list[4])

df_train = pd.DataFrame()
df_test = pd.DataFrame()
for each in json_list[:-1]:
    if 'aAcc' in each:
        df_test = df_test.append(eval(each), ignore_index=True)
    else:
        df_train = df_train.append(eval(each), ignore_index=True)

导出训练日志表格

df_train.to_csv('训练日志-训练集.csv', index=False)
df_test.to_csv('训练日志-测试集.csv', index=False)

可视化辅助函数

from matplotlib import colors as mcolors
import random
random.seed(124)
colors = ['b', 'g', 'r', 'c', 'm', 'y', 'k', 'tab:blue', 'tab:orange', 'tab:green', 'tab:red', 'tab:purple', 'tab:brown', 'tab:pink', 'tab:gray', 'tab:olive', 'tab:cyan', 'black', 'indianred', 'brown', 'firebrick', 'maroon', 'darkred', 'red', 'sienna', 'chocolate', 'yellow', 'olivedrab', 'yellowgreen', 'darkolivegreen', 'forestgreen', 'limegreen', 'darkgreen', 'green', 'lime', 'seagreen', 'mediumseagreen', 'darkslategray', 'darkslategrey', 'teal', 'darkcyan', 'dodgerblue', 'navy', 'darkblue', 'mediumblue', 'blue', 'slateblue', 'darkslateblue', 'mediumslateblue', 'mediumpurple', 'rebeccapurple', 'blueviolet', 'indigo', 'darkorchid', 'darkviolet', 'mediumorchid', 'purple', 'darkmagenta', 'fuchsia', 'magenta', 'orchid', 'mediumvioletred', 'deeppink', 'hotpink']
markers = [".",",","o","v","^","<",">","1","2","3","4","8","s","p","P","*","h","H","+","x","X","D","d","|","_",0,1,2,3,4,5,6,7,8,9,10,11]
linestyle = ['--', '-.', '-']

def get_line_arg():
    '''
    随机产生一种绘图线型
    '''
    line_arg = {}
    line_arg['color'] = random.choice(colors)
    # line_arg['marker'] = random.choice(markers)
    line_arg['linestyle'] = random.choice(linestyle)
    line_arg['linewidth'] = random.randint(1, 4)
    # line_arg['markersize'] = random.randint(3, 5)
    return line_arg

训练集损失函数

metrics = ['loss', 'decode.loss_ce', 'aux.loss_ce']
plt.figure(figsize=(16, 8))

x = df_train['step']
for y in metrics:
    plt.plot(x, df_train[y], label=y, **get_line_arg())

plt.tick_params(labelsize=20)
plt.xlabel('step', fontsize=20)
plt.ylabel('loss', fontsize=20)
plt.title('训练集损失函数', fontsize=25)
plt.savefig('训练集损失函数.pdf', dpi=120, bbox_inches='tight')

plt.legend(fontsize=20)

plt.show()

在这里插入图片描述

训练集准确率

plt.figure(figsize=(16, 8))

x = df_train['step']
for y in metrics:
    plt.plot(x, df_train[y], label=y, **get_line_arg())

plt.tick_params(labelsize=20)
plt.xlabel('step', fontsize=20)
plt.ylabel('loss', fontsize=20)
plt.title('训练集准确率', fontsize=25)
plt.savefig('训练集准确率.pdf', dpi=120, bbox_inches='tight')

plt.legend(fontsize=20)

plt.show()

在这里插入图片描述

测试集评估指标

plt.figure(figsize=(16, 8))

x = df_test['step']
for y in metrics:
    plt.plot(x, df_test[y], label=y, **get_line_arg())

plt.tick_params(labelsize=20)
plt.ylim([0, 100])
plt.xlabel('step', fontsize=20)
plt.ylabel(y, fontsize=20)
plt.title('测试集评估指标', fontsize=25)
plt.savefig('测试集分类评估指标.pdf', dpi=120, bbox_inches='tight')

plt.legend(fontsize=20)

plt.show()

在这里插入图片描述

模型预测

导入工具包

import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline

from mmseg.apis import init_model, inference_model, show_result_pyplot
import mmcv
import cv2

载入配置文件

# 载入 config 配置文件
from mmengine import Config
cfg = Config.fromfile('pspnet-DubaiDataset_20230612.py')
from mmengine.runner import Runner
from mmseg.utils import register_all_modules

# register all modules in mmseg into the registries
# do not init the default scope here because it will be init in the runner

register_all_modules(init_default_scope=False)
runner = Runner.from_cfg(cfg)

载入模型

checkpoint_path = './work_dirs/DubaiDataset/iter_3000.pth'
model = init_model(cfg, checkpoint_path, 'cuda:0')

载入测试集图像，或新图像

img = mmcv.imread('Dubai-dataset/img_dir/val/71.jpg')

语义分割预测

result = inference_model(model, img)
pred_mask = result.pred_sem_seg.data[0].cpu().numpy()
np.unique(pred_mask)
array([0, 1, 2, 3, 4])
plt.imshow(pred_mask)
plt.show()

在这里插入图片描述

可视化预测结果

visualization = show_result_pyplot(model, img, result, opacity=0.7, out_file='pred.jpg')
plt.imshow(mmcv.bgr2rgb(visualization))
plt.show()

在这里插入图片描述

获取测试集标注

获取测试集标注
label = mmcv.imread('Dubai-dataset/ann_dir/val/71.png')
label_mask = label[:,:,0]
np.unique(label_mask)
plt.imshow(label_mask)
plt.show()

在这里插入图片描述

对比测试集标注和语义分割预测结果

# 测试集标注

# 真实为前景，预测为前景
TP = (label_mask == 1) & (pred_mask==1)
# 真实为背景，预测为背景
TN = (label_mask == 0) & (pred_mask==0)
# 真实为前景，预测为背景
FN = (label_mask == 1) & (pred_mask==0)
# 真实为背景，预测为前景
FP = (label_mask == 0) & (pred_mask==1)
plt.imshow(TP)
plt.show()

在这里插入图片描述

confusion_map = TP * 255 + FP * 150 + FN * 80 + TN * 30
plt.imshow(confusion_map)
plt.show()

在这里插入图片描述
混淆矩阵

from sklearn.metrics import confusion_matrix
confusion_matrix_model = confusion_matrix(label_mask.flatten(), pred_mask.flatten())
confusion_matrix_model

import itertools
def cnf_matrix_plotter(cm, classes, cmap=plt.cm.Blues):
    """
    传入混淆矩阵和标签名称列表，绘制混淆矩阵
    """
    plt.figure(figsize=(10, 10))
    
    plt.imshow(cm, interpolation='nearest', cmap=cmap)
    # plt.colorbar() # 色条
    tick_marks = np.arange(len(classes))
    
    plt.title('Confusion Matrix', fontsize=30)
    plt.xlabel('Pred', fontsize=25, c='r')
    plt.ylabel('True', fontsize=25, c='r')
    plt.tick_params(labelsize=16) # 设置类别文字大小
    plt.xticks(tick_marks, classes, rotation=90) # 横轴文字旋转
    plt.yticks(tick_marks, classes)
    
    # 写数字
    threshold = cm.max() / 2.
    for i, j in itertools.product(range(cm.shape[0]), range(cm.shape[1])):
        plt.text(j, i, cm[i, j],
                 horizontalalignment="center",
                 color="white" if cm[i, j] > threshold else "black",
                 fontsize=12)

    plt.tight_layout()

    plt.savefig('混淆矩阵.pdf', dpi=300) # 保存图像
    plt.show()
classes = ['Land', 'Road', 'Building', 'Vegetation', 'Water', 'Unlabeled']
cnf_matrix_plotter(confusion_matrix_model, classes, cmap='Blues')

在这里插入图片描述

测试集性能评估

测试集精度指标

python tools/test.py pspnet-DubaiDataset_20230612.py work_dirs/DubaiDataset/iter_3000.pth

速度指标-FPS

!python tools/analysis_tools/benchmark.py pspnet-DubaiDataset_20230612.py work_dirs/DubaiDataset/iter_3000.pth

Kaggle实战-小鼠肾小球组织病理切片语义分割

下载整理好的数据集

下载数据集

!rm -rf Glomeruli-dataset.zip Glomeruli-dataset
!wget https://zihao-openmmlab.obs.cn-east-3.myhuaweicloud.com/20230130-mmseg/dataset/Glomeruli-dataset.zip
!unzip Glomeruli-dataset.zip >> /dev/null # 解压
!rm -rf Glomeruli-dataset.zip # 删除压缩包

删除系统自动生成的多余文件
查看待删除的多余文件

!find . -iname '__MACOSX'
!find . -iname '.DS_Store'
!find . -iname '.ipynb_checkpoints'
./.ipynb_checkpoints
./Glomeruli-dataset/.ipynb_checkpoints
./Glomeruli-dataset/splits/.ipynb_checkpoints

删除多余文件

!for i in `find . -iname '__MACOSX'`; do rm -rf $i;done
!for i in `find . -iname '.DS_Store'`; do rm -rf $i;done
!for i in `find . -iname '.ipynb_checkpoints'`; do rm -rf $i;done

验证多余文件已删除

!find . -iname '__MACOSX'
!find . -iname '.DS_Store'
!find . -iname '.ipynb_checkpoints'

探索数据集

导入工具包

import os

import cv2
import numpy as np
from PIL import Image
from tqdm import tqdm

import matplotlib.pyplot as plt
%matplotlib inline

指定图像和标注文件夹路径

PATH_IMAGE = 'Glomeruli-dataset/images'
PATH_MASKS = 'Glomeruli-dataset/masks'
print('图像个数', len(os.listdir(PATH_IMAGE)))
print('标注个数', len(os.listdir(PATH_MASKS)))

查看单张图像及其语义分割标注

指定图像文件名

file_name = 'SAS_21883_001_10.png'
img_path = os.path.join(PATH_IMAGE, file_name)
mask_path = os.path.join(PATH_MASKS, file_name)

print('图像路径', img_path)
print('标注路径', mask_path)
img = cv2.imread(img_path)
mask = cv2.imread(mask_path)

# 可视化图像
plt.imshow(img)
plt.show()

在这里插入图片描述

# mask 语义分割标注，与原图大小相同，0 为 背景， 1 为 肾小球
np.unique(mask)

array([0, 1], dtype=uint8)
在本数据集中，只有一部分图像有肾小球语义分割标注（即mask中值为1的像素），其余图像mask的值均为0

可视化语义分割标注

plt.imshow(mask[:,:,0])
plt.show()

# 可视化语义分割标注
plt.imshow(mask*255)
plt.show()

在这里插入图片描述

可视化单张图像及其语义分割标注-代码模板

plt.imshow(img)
plt.imshow(mask*255, alpha=0.5) # alpha 高亮区域透明度，越小越接近原图
plt.title(file_name)
plt.axis('off')
plt.show()

在这里插入图片描述

可视化模板-有前景标注

# n行n列可视化
n = 7

# 标注区域透明度
opacity = 0.5

fig, axes = plt.subplots(nrows=n, ncols=n, sharex=True, figsize=(12,12))

i = 0

for file_name in os.listdir(PATH_IMAGE):
    
    # 载入图像和标注
    img_path = os.path.join(PATH_IMAGE, file_name)
    mask_path = os.path.join(PATH_MASKS, file_name)
    img = cv2.imread(img_path)
    mask = cv2.imread(mask_path)
    
    if 1 in mask:
        axes[i//n, i%n].imshow(img)
        axes[i//n, i%n].imshow(mask*255, alpha=opacity)
        axes[i//n, i%n].axis('off') # 关闭坐标轴显示
        i += 1
    if i > n**2-1:
        break
fig.suptitle('Image and Semantic Label', fontsize=30)
plt.tight_layout()
plt.show()

在这里插入图片描述

可视化模板-无论前景是否有标注

# n行n列可视化
n = 10

# 标注区域透明度
opacity = 0.5

fig, axes = plt.subplots(nrows=n, ncols=n, sharex=True, figsize=(12,12))

for i, file_name in enumerate(os.listdir(PATH_IMAGE)[:n**2]):
    
    # 载入图像和标注
    img_path = os.path.join(PATH_IMAGE, file_name)
    mask_path = os.path.join(PATH_MASKS, file_name)
    img = cv2.imread(img_path)
    mask = cv2.imread(mask_path)
    
    # 可视化
    axes[i//n, i%n].imshow(img)
    axes[i//n, i%n].imshow(mask*255, alpha=opacity)
    axes[i//n, i%n].axis('off') # 关闭坐标轴显示
fig.suptitle('Image and Semantic Label', fontsize=30)
plt.tight_layout()
plt.show()