Awesome
Camouflaged Object Segmentation with Prior via Two-stage Training
This project provides the code and results for 'Camouflaged Object Segmentation with Prior via Two-stage Training'
Authors: Rui Wang Caijuan Shi, Changyu Duan, Weixiang Gao, Hongli Zhu, Yunchao Wei
Network Architecture
Preparation
The training and testing experiments are conducted using PyTorch with a single GeForce RTX 1080Ti GPU of 12 GB Memory.
Configuring your environment:
- Creating a virtual environment :
conda create -n SINet python=3.9
- Installing necessary packages:
pip install -r requirements.txt
Downloading Training and Testing Sets
- Download train datasets (COD10K-train+CAMO-train): TrainDatasets
- Download test datasets (CAMO-test+COD10K-test-test+CHAMELEON+NC4K ):TestDatasets
Pretrained Backbone Model
- Download pretrained backbone model:PVTv2-b4, and put it in
./pth
Training
- Modify the dataset path in
config.py --freeze=decoder
. - First Training: run
python train.py
, and it generates catalogueexperiments\
with logs and weights. - Second Training: run
python train.py --ckpt=last --freeze=backbone --thaw=backbone
- You can also change the other config option by modify the
config.py
.
Testing Configuration
- Testing: run
python test.py
, and the result maps are inexperiments\save_images\
. - We provide CGTNet testing maps and training weights presented in the papers.
Evaluation
- Tools: PySODMetrics A simple and efficient implementation of SOD metrics.
Results
Credit
The code is partly based on Camoformer.