Unofficial Implementation due to confidentiality matters.
All the codes are written under PyTorch framework, and are designed for one or more GPUs acceleration.
The pretrained weights for the saliency detection network can be found on PyTorch Hub, but they are not fine-tuned specifically for. By the way, this network has been designed to respect the architecture of those same weights, so you can just load them and run the code.