Roudgers / DCFNet

Dynamic Context-Sensitive Filtering Network for Video Salient Object Detection

DCFNet

Accepted paper in ICCV2021:

Dynamic Context-Sensitive Filtering Network for Video Salient Object Detection

Miao Zhang, Jie Liu, Yifei Wang, Yongri Piao, Shunyu Yao, Wei Ji, Jingjing Li, Huchuan Lu, Zhongxuan Luo.

Prerequisites

• Ubuntu 16
• PyTorch 1.6.0
• CUDA 10.1
• Cudnn 7.5.0
• Python 3.6

VSOD Training and Testing Datasets

VSOD Training dataset

VSOD Training dataset. Code: oip1

VSOD Testing dataset

VSOD Testing dataset. Code: oip1

Train/Test

Testing

• Firstly, you need to download the 'VSOD Testing Dataset' and the pretrained checkpoint we provided (video_current_best_model.pth. Code: oip1).
• Secondly, you need to set dataset path and checkpoint name correctly, and set the param '--split' as "test" or 'val' in inference.py for generating saliency results.
• Finally, you can evaluate the saliency results by using the widely-used tool provided by DAVSOD.
• Or you can directly download the results we provided (DCFNet results. Code: oip1).

python inference.py

Training

• Firstly, you need to download the 'VSOD Training Dataset' and modify your path of training dataset
• Secondly, you can pretrain the ImageModel for DCFNet following the training settings in the paper
• Finally, you can train the VideoModel for DCFNet

python train.py

Contact Us

If you have any questions, please contact us ([e-mail]; [e-mail]).

Acknowledge

Thanks to the previous helpful works:

• RCRNet: 'Semi-Supervised Video Salient Object Detection Using Pseudo-Labels' by Pengxiang Yan, et al.
• SSAV: 'Shifting More Attention to Video Salient Object Detection' by Deng-Ping Fan, et al.