zsbluesky/saldetect

The code for the joint learning of saliency prediction and object detection (using Faster RCNN).

Usage

Please copy lib dir in faster_rcnn.pytorch to your root dir and install faster_rcnn.pytorch, following the Preparation steps. Prepare the dataset of Pascal VOC according to faster_rcnn.pytorch. Please prepare SALICON dataset according to EML-Net.

Training (Joint Learning) and Testing (Saliency Prediction)

Please refer to train2.py and test_net.py. Change the predefined parameters according to your own data path, model path, etc. Here we do the joint learning of object detection and saliency prediction. While in the test stage, we only output saliency result. We use the Resnet50 backbone in the code.

Acknowledgement

The code is based on faster_rcnn.pytorch and EML-Net. Hope the code is helpful for the joint learning in the field of saliency prediction.