AGSS

This is the demo code for ''Adapting Generic RGB-D Salient Object Detection for Specific Scenarios''.

Getting Started

Requirements

• Python 3.7+
• Pytorch 1.6.0+
• CUDA v10.1, cudnn v.7.5.0
• torchvision

Data Preprocessing

Download the RGB-D training datasets and testing datasets from dataset1 Code: qj71 and dataset2 Code: ehh8.

Download the stereo training datasets and testing datasets from datasets Code: 8nrw.

Download our collected Video8K from dataset1 Code: fvfd and dataset2 Code: rkmr.

Step Warming up

Generate optical flow maps (refer to ./GenOpticalFlow/)

Step 1

Generate saliency informative depth (refer to ./GenDepth/)

Step 2

Generate high-quality pseudo-GT (refer to ./GenPseudoGT/)

Step 3

Target Domain Adaption (refer to ./Target Models/)

• Modify your path of testing dataset in test.py
• Run test.py to inference saliency maps
• Saliency maps on 9 RGB-D testing datasets and 2 stereo testing datasets generated from the three target models can be downnloaded from here Code: awc4.

Acknowledgement

Thanks to SPNet, SSL and C2DFNet.