赞
踩
参考我上一篇文章:ubuntu编译安装mmcv 1.6.2和mmsegmentation 0.28.0
参考:https://github.com/open-mmlab/mmsegmentation/blob/master/docs/zh_cn/dataset_prepare.md
gtFine_trainvaltest.zip (241MB)
、gtCoarse.zip (1.3GB)
、leftImg8bit_trainvaltest.zip (11GB)
mmsegmentation
├── mmseg
├── tools
├── configs
├── data
│ ├── cityscapes
│ │ ├── leftImg8bit
│ │ │ ├── train
│ │ │ ├── val
│ │ ├── gtFine
│ │ │ ├── train
│ │ │ ├── val
│ │ ├── gtCoarse
│ │ │ ├── train
│ │ │ ├── train_extra
│ │ │ ├── val
# 单卡GPU 测试
python ./tools/test.py configs/segformer/segformer_mit-b0_8x1_1024x1024_160k_cityscapes.py ./checkpoints/segformer_mit-b0_8x1_1024x1024_160k_cityscapes_20211208_101857-e7f88502.pth --eval cityscapes --show
# 多卡GPU 测试
./tools/dist_test.sh configs/segformer/segformer_mit-b0_8x1_1024x1024_160k_cityscapes.py ./checkpoints/segformer_mit-b0_8x1_1024x1024_160k_cityscapes_20211208_101857-e7f88502.pth 1 --eval cityscapes --show
*注: 多卡GPU测试中1
为GPU数目
python tools/train.py configs/segformer/segformer_mit-b5_8x1_1024x1024_160k_cityscapes.py
默认会生成work_dir存放训练过程中的log和checkpoints:
输出:
Copyright © 2003-2013 www.wpsshop.cn 版权所有,并保留所有权利。