"Adapting Transformer in Image Segmentation tasks"
2020021461 Heejo Kong
1st project for Applications and Practice in Neural Networks
This repository is based on official code of mmsegmentation & SegFormer
https://github.com/open-mmlab/mmsegmentation
https://github.com/NVlabs/SegFormer
Installation
For install and data preparation, please refer to the guidelines as follows
pip install torchvision==0.8.2
pip install timm==0.3.2
pip install mmcv-full==1.2.7 -f https://download.openmmlab.com/mmcv/dist/{cuda version}/{torch version}/index.html
# example, https://download.openmmlab.com/mmcv/dist/cu101/torch1.7.0/index.html
pip install opencv-python==4.5.1.48
git clone https://github.com/deepkong/NN_project1_kong.git
cd NN_project1_kong && pip install -e .
Donwload [dataset] https://drive.google.com/file/d/1TTFloQwNdg95h-IJceUtVs1WQl_Bu1Ev/view?usp=sharing
Need to prepare the data root as below
NN_project1_kong
├── mmseg
├── tools
├── local_configs
├── data
│ ├── semantic_drone_dataset
│ │ ├── images
│ │ ├── labels
│ │ ├── split
│ │ │ ├── train.txt
│ │ │ ├── val.txt
│ │ │ ├── test.txt
Training
Donwload [pretrained weights] https://drive.google.com/file/d/1VuhtdeQSGl2gxndg_-98HPFJ1qEl3KiH/view?usp=sharing
# Single-gpu training
python tools/train.py local_configs/segformer_drone/segformer.b2.512x512.drone.20k.py
# Multi-gpu training
./tools/dist_train.sh local_configs/segformer_drone/segformer.b2.512x512.drone.20k.py <GPU_NUM>
Evaluation
Donwload [weights] https://drive.google.com/file/d/15X00UIMRd3cmFC7sgKiHu5DCRQhM5zik/view?usp=sharing
# Single-gpu testing
python tools/test.py local_configs/segformer_drone/segformer.b2.512x512.drone.20k.py /path/to/checkpoint_file --show-dir /path/to/save_file
# Multi-gpu testing
./tools/dist_test.sh local_configs/segformer_drone/segformer.b2.512x512.drone.20k.py /path/to/checkpoint_file <GPU_NUM> --show-dir /path/to/save_file
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent