(CVPR 2022) Zoom In and Out: A Mixed-scale Triplet Network for Camouflaged Object Detection

@inproceedings{ZoomNet-CVPR2022,
	title     = {Zoom In and Out: A Mixed-scale Triplet Network for Camouflaged Object Detection},
	author    = {Pang, Youwei and Zhao, Xiaoqi and Xiang, Tian-Zhu and Zhang, Lihe and Lu, Huchuan},
	booktitle = CVPR,
	year      = {2022}
}

Changelog

2022-3-16
- Add the link of the method prediction maps of Table 1 in our paper.
2022-03-08
- Add the link of arxiv version.
2022-03-07
- Add the link of paper.
2022-03-05:
- Update weights and results links.
- Fixed some bugs.
- Update dataset links.
- Update bibtex info.
2022-03-04:
- Initialize the repository.
- Add the model and configuration file for SOD.

Usage

Dependencies

Some core dependencies:

timm == 0.4.12
torch == 1.8.1
pysodmetrics == 1.2.4 # for evaluating results

More details can be found in <./requirements.txt>

Datasets

More details can be found at:

Training

You can use our default configuration, like this:

$ python main.py --model-name=ZoomNet --config=configs/zoomnet/zoomnet.py --datasets-info ./configs/_base_/dataset/dataset_configs.json --info demo

or use our launcher script to start the one command in commands.txt on GPU 1:

$ python tools/run_it.py --interpreter 'abs_path' --cmd-pool tools/commands.txt  --gpu-pool 1 --verbose --max-workers 1

If you want to launch multiple commands, you can use it like this:

Add your commands into the tools/commands.txt.
python tools/run_it.py --interpreter 'abs_path' --cmd-pool tools/commands.txt --gpu-pool <gpu indices> --verbose --max-workers max_workers

NOTE:

abs_path: the absolute path of your python interpreter
max_workers: the maximum number of tasks to start simultaneously.

Testing

Task	Weights	Results
COD	GitHub Release Link	GitHub Release Link
SOD	GitHub Release Link	GitHub Release Link

For ease of use, we create a test.py script and a use case in the form of a shell script test.sh.

$ sudo chmod +x ./test.sh
$ ./test.sh 0  # on gpu 0

Method Comparisons

The prediction maps corresponding to the methods in Table 1 of our paper:
- Baidu Pan: https://pan.baidu.com/s/1dLMqa4tix1gdBN1uWrCPbQ Code: yxy9
PySODEvalToolkit: A Python-based Evaluation Toolbox for Salient Object Detection and Camouflaged Object Detection
- https://github.com/lartpang/PySODEvalToolkit