DSNet: Dynamic Spatiotemporal Network for Video Salient Object Detection
-
Download pretrained ResNet34 to './model/resnet/pre_train/'.
-
Organize each dataset according to the organization format in the './dataset/DAVIS'.
-
Install FlowNet2.0 and genrate correct optical flow images.
-
Start to train with
python main.py --mode train
.
-
Download pretrained model (87xl) to './model/pretrained/'
-
Generate saliency maps for SOD dataset by
python main.py --mode test
.
-
The saliency maps can be download in https://pan.baidu.com/s/1snIniA836GEtIvCSjhE2FQ(5a6m)
-
The evaluation tools are provided by Fan D et. al.