Awesome

MCNet

SwinMCNet: Mirror Complementary Transformer Network for RGB-thermal Salient Object Detection

Prerequisites

• Python 3.68
• Pytorch 1.3.1
• Cuda 10.0
• OpenCV 4.1.2
• Numpy 1.17.3
• TensorboardX 2.1

Benchmark Datasets

Download the following datasets and unzip them into data folder

• VT5000
• VT1000
• VT821

The Proposed Dataset

Our proposed RGBT SOD dataset VT723 that contain common challenging scenes of real world.

• VT723

Training & Testing & Evaluate

• Split the ground truth into skeleton map and contour map, which will be saved into data/VT5000/skeleton and data/VT5000/contour.

    python3 utils.py

• Train the model and get the pretrained model, which will be saved into res folder.

    python3 train.py

• If you just want to evaluate the performance of MCNet without training, please download the pretrained model into res folder.
• Test the model and get the predicted saliency maps, which will be saved into maps folder.

   python3 test.py

• Evaluate the predicted results.

    cd eval
    matlab
    main

Saliency maps & Trained model

• saliency maps: Google
• trained model: Google