Awesome
BiconNets
BiconNet: An Edge-preserved Connectivity-based Approach for Salient Object Detection
Requirement: Pytorch 1.7.1
This code including three parts:
- Codes for customizing BiconNet wtih other backbones (/general)
- Codes for reproducing the paper results (/paper_result)
- Evaluation Code (/evaluation)
- Customize the BiconNet based on your own network. (/general) If you want to construct the BiconNet based on your own network, there are four simple steps:
- replace your network's one-channel output fully connected layers with 8-channel FC layers.
For training: 2) generate the ground truth connectivity masks using the function 'sal2conn' in utils_bicon.py 3) replace your own loss function with Bicon_loss: you can edit the connect_loss.py
For testing: 4) use the function 'bv_test' in utils_bicon.py after you get the 8-channel connectivity map output to get your final saliency prediction. 2. Customize the BiconNet based on your own network. (/general) If you want to construct the BiconNet based on your own network, there are four simple steps:
- replace your network's one-channel output fully connected layers with 8-channel FC layers.
For training: 2) generate the ground truth connectivity masks using the function 'sal2conn' in utils_bicon.py 3) replace your own loss function with Bicon_loss: you can edit the connect_loss.py
For testing: 4) use the function 'bv_test' in utils_bicon.py after you get the 8-channel connectivity map output to get your final saliency prediction.
- Reproduce the results in the paper (/paper_result)
(a) PoolNet Baseline code from: https://github.com/backseason/PoolNet
For traing: cd /PoolNet/bicon/train python train.py
For testing: cd /PoolNet/bicon/test python test.py
make sure the datapath is correct.
(b) CPD-R Baseline code from: https://github.com/wuzhe71/CPD
For training cd /CPD-R/bicon/train python train.py
For testing: cd /CPD-R/bicon/test python test.py
make sure the datapath is correct.
(c) EGNet Baseline code from: https://github.com/JXingZhao/EGNet
For training cd /EGNet/bicon/train python run.py
For testing: cd /EGNet/bicon/test python test.py
make sure the datapath is correct.
(d) GCPANet Baseline code from: https://github.com/JosephChenHub/GCPANet
For training cd /GCPANet/bicon/train python train.py
For testing: cd /GCPANet/bicon/test python test.py
make sure the datapath is correct.
(d) ITSD Baseline code from: https://github.com/moothes/ITSD-pytorch
For training cd /ITSD/bicon/train python train.py
For testing: cd /ITSD/bicon/test python test.py
make sure the datapath is correct.
(d) MINet Baseline code from: https://github.com/lartpang/MINet
For training cd /MINet/bicon/train python main.py
For testing: cd /MINet/bicon/test python main.py
make sure the datapath is correct.
- Results evaluation (/evaluation) We use Matlab to evaluate the output saliency maps as did in: https://github.com/JosephChenHub/GCPANet