DS-Net

DSNet: Dynamic Spatiotemporal Network for Video Salient Object Detection

For training:

1. Download pretrained ResNet34 to './model/resnet/pre_train/'.

2. Organize each dataset according to the organization format in the './dataset/DAVIS'.

3. Install FlowNet2.0 and genrate correct optical flow images.

4. Start to train with python main.py --mode train.

For testing:

1. Download pretrained model (87xl) to './model/pretrained/'

2. Generate saliency maps for SOD dataset by python main.py --mode test.

For evaluation:

1. The saliency maps can be download in https://pan.baidu.com/s/1snIniA836GEtIvCSjhE2FQ(5a6m)

2. The evaluation tools are provided by Fan D et. al.