DHS net

DHS net for salient objects detection(Pytorch implementation). Some part of this project is based on codes from https://github.com/wlguan/DHSNet-PyTorch , and this project is an optimized version.

Requirements

Original running environment:

• Python 3.7.5
• Pytorch 1.3.1
• TorchVision 0.2.1
• pillow 7.0.0

See requirements.txt for detail.

Training

1. Put corresponding dataset in ./input/
   • training images(RGB, jpg format): ./input/train/raw/
   • training masks(gray, png format): ./input/train/mask/
   • validation images(RGB, jpg format): ./input/test/raw/
   • validation masks(gray, png format): ./input/test/mask/
2. Run train.py, if you want to change some parameters, see train.py for detail.

Inference

1. Put inference data in ./inference/
   • inference images(RGB, jpg format): ./inference
2. Run inference.py, output saliency maps will be in ./output directory.