DSLRDNet

DSLRDNet: Addressing Multiple Salient Object Detection via Dual-Space Long-Range Dependencies

Prerequisites:

1. Pytorch 1.2.0
2. Opencv 2.4.5
3. TensorboardX

For training:

1. Download the DUTS-TR (Google Drive) training dataset.
2. Download the initial pratrained VGG/ResNet (Google Drive) model.
3. Change the training data path in dataset.py.
4. Change the training settings in solver.py and run.py
5. Start to train with python3 run.py --mode train

For testing:

1. Download the pretrained models (UoN server).
2. Change the data path in dataset.py
3. Change the test settings in run.py.
4. Generate saliency maps with python3 run.py --mode test --sal_mode m, where 'm' demonstrates the MSOD dataset.
5. We use the public open source evaluation code. (https://github.com/weijun88/F3Net)

Datasets and results:

MSOD dataset || Generated Saliency Maps