Prompt-Driven Multi-Task Learning with Task Tokens for ORSI Salient Object Detection

Welcome to the official repository for the paper "Prompt-Driven Multi-Task Learning with Task Tokens for ORSI Salient Object Detection".

Network Architecture

Motivate

Utilize task prompts to enforce orthogonality between the gradient directions of two tasks.

Comparison with SOTA methods

The Initialization Weights for Training

Download pre-trained classification weights of the Swin Transformer , and place the .pth files in ./pretrained_model directory. These weights are essential for initializing the model during training.

Trained Weights of MTPNet for Testing

We provide Trained Weights of our MTPNet. Download

Train

Please download the pre-trained model weights and dataset first. Next, generate the path of the training set and the test set, and change the dataset path in the code to the path of the dataset you specified.

python train.py

Test

Download the MTPNet model weights, create the necessary directories to store these files, and be sure to update the corresponding paths in the code accordingly.

python test.py

Saliency maps

We provide saliency maps of our MTPNet on ORSSD，EORSSD and ORSI-4199 datasets. Download

Evaluation Tool

You can use the evaluation tool (MATLAB version) to evaluate the above saliency maps.