The equisym from ahyunseo

equisym's Introduction

Reflection and Rotation Detection via Equivariant Learning (CVPR 2022)

Ahyun Seo, Byungjin Kim, Suha Kwak, Minsu Cho

Official PyTorch implementation of Reflection and Rotation Detection via Equivariant Learning (CVPR 2022).

Environment

    conda create --name EquiSym python=3.7
    conda activate EquiSym
    conda install pytorch==1.7.0 torchvision==0.8.1 cudatoolkit=11.0 -c pytorch
    conda install -c conda-forge matplotlib
    pip install albumentations==0.5.2 shapely opencv-python tqdm e2cnn mmcv
    
    mkdir weights wandb sym_datasets

Datasets and weights

download DENDI onedrive or DENDI
trained weights: EquiSym(ours), EquiSym(CNN ver.), pre-trained ReResNet50(D8)

.
├── sym_datasets
│   └── DENDI
│       ├── symmetry
│       ├── symmetry_polygon
│       ├── reflection_split.pt
│       ├── rotation_split.pt
│       └── joint_split.pt
├── weights
│   ├── v_equiv_aux_ref_best_checkpoint.pt
│   ├── v_equiv_aux_rot_best_checkpoint.pt
│   ├── v_cnn_ref_best_checkpoint.pt
│   ├── v_cnn_rot_best_checkpoint.pt
│   └── re_resnet50_custom_d8_batch_512.pth
├── (...) 
└── main.py

Demo & Test

visualize results using the input images in ./imgs

    python demo.py --ver equiv_aux_ref -rot 0 -eq --get_theta 10  
    python demo.py --ver equiv_aux_rot -rot 1 -eq --get_theta 10

test with pretrained weights

    python train.py --ver equiv_aux_ref -t -rot 0 -eq -wf --get_theta 10 
    python train.py --ver equiv_aux_rot -t -rot 1 -eq -wf --get_theta 10

vis(test) with pretrained weights of vanilla CNN model

    python demo.py --ver cnn_ref -rot 0
    python demo.py --ver cnn_rot -rot 1

Training

The trained weights and arguments will be save to the checkpoint path corresponding to the VERSION_NAME.

    python train.py --ver VERSION_NAME_REF -tlw 0.01 --get_theta 10 -rot 0 -eq
    python train.py --ver VERSION_NAME_ROT -tlw 0.001 --get_theta 10 -rot 1 -eq

References

ReDet link
e2cnn link
pytorch-deeplab-xception link
PMCNet link

Citation

If you find our code or paper useful to your research work, please consider citing:

@inproceedings{seo2022equisym,
    author   = {Seo, Ahyun and Kim, Byungjin and Kwak, Suha and Cho, Minsu},
    title    = {Reflection and Rotation Symmetry Detection via Equivariant Learning},
    booktitle= {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
    year     = {2022}
}

equisym's People

Contributors

Stargazers

Watchers

equisym's Issues

wandb.errors.CommError

wandb.errors.CommError: It appears that you do not have permission to access the requested resource. Please reach out to the project owner to grant you access. If you have the correct permissions, verify that there are no issues with your networking setup.(Error 404: Not Found)
请问作者大大这个是不是要您给权限才能运行？现在您的代码可以让我自己训练权重吗？

Can models be used to train their own datasets?

How did you get the format of the dataset?

下载mmcv报错

ERROR: Command errored out with exit status 1:
command: /usr/bin/python -c 'import sys, setuptools, tokenize; sys.argv[0] = '"'"'/tmp/pip-install-fF44V0/mmcv/setup.py'"'"'; file='"'"'/tmp/pip-install-fF44V0/mmcv/setup.py'"'"';f=getattr(tokenize, '"'"'open'"'"', open)(file);code=f.read().replace('"'"'\r\n'"'"', '"'"'\n'"'"');f.close();exec(compile(code, file, '"'"'exec'"'"'))' egg_info --egg-base /tmp/pip-pip-egg-info-tungav
cwd: /tmp/pip-install-fF44V0/mmcv/
Complete output (6 lines):
Traceback (most recent call last):
File "", line 1, in
File "/tmp/pip-install-fF44V0/mmcv/setup.py", line 101
yield from parse_line(line)
^
SyntaxError: invalid syntax
----------------------------------------
ERROR: Command errored out with exit status 1: python setup.py egg_info Check the logs for full command output.
下载mmcv报错，想问下佬遇到过这个问题吗

Cannot run demo

ImportError: cannot import name 'constant_init' from 'mmcv.cnn'

I followed the installation directions. mmcv 2.1.0 is installed. I do not have CUDA on my system.

Recommend Projects

ahyunseo / equisym Goto Github PK