This repo. is an implementation of R2Net, which is accepted for in Image and Vision Computing.
The paper is here.
you can find the saliency maps on DUTS-TE、ECSSD、HKU-IS、DUT-OMRON and PASCAL-S datasets and the weight file from Google Driver link and the Baidu online disk link (Code:RRNe)
The train.py file contains the training code, If you want to retrain, please download the training set and test set from here, and unzip the file, then modify the "train_dataset" parameter to your own path.
if you want to test, just modify the path of the saliency maps, and I think you have your own testing code. So I provide our saliency maps additionally.
We use the code provided by this repo. to calculate the metrics.
We choose PaddlePaddle as the framework, in particular, PaddlePaddle provides a good learning environment and hardware facilities. If you want the PyTorch version of the code, it will be available right away.
The effect of R^2Net on 5 benchmark datasets is as follows, we achieve the SOTA results than any existing SOD methods.