Co-saliency object detection with FCN

Our caffemodel file can be downloaded here

Our Environment: Ubuntu 18.04, CUDA9.0

We use Image Label Data Layer from HED Caffe, so if you are using official Caffe please add the layer manually and re-compile the Caffe. Or you can directly install this version.

The usage of some files are here below

• run_saliency.py: Train the model, please alter the location of base model before running
• saliency_test.py: Test the model. You can either print the images on the screen or directly save the output images.
• mul.py & CRF.py: Further processing the output images from FCN (saliency_test) with method A and B in the paper respectively. Notice: mul.py requires user generate the single-image saliency map before (we use method DSS ). CRF.py we use pyDense, please install it first.
• comp_.py: Compare results with different further improvement like CRF, mul
• some other python files are used to generate image lists, you can read them if you need them.

Notice, please check the data path in the code before using.