CATNet: A Cascaded and Aggreated Vision Transformer Network for RGB-D Salient Object Detection

Network Architecture

Requirements

python 3.9

pytorch 1.11.0

tensorboardX 2.5

Results and Saliency maps

We provide saliency maps(baidu (code:) | Google Drive) of our CATNet on 7 datasets. We Perform Quantitative Comparisons With CNN-Based RGB-D SOD Methods On Seven RGB-D Datasets. We Perform Quantitative Comparisons With Transformer-Based RGB-D SOD Methods On Seven RGB-D Datasets.

Training

Please run

CatNet_train.py

Pre-trained model and testing

• Download the following pre-trained models and put them in /pretrain.
• Modify pathes of pre-trained models and datasets.
• Run

CatNet_test.py

Datasets: Baidu(code:) | Google Drive

Our pre-trained models(CATNet): Baidu(code:) | Google Drive