Skip to content
/ OBGNet Public

code for ACM MM 2021 paper "Occlusion-aware Bi-directional Guided Network for Light Field Salient Object Detection"

Notifications You must be signed in to change notification settings

Timsty1/OBGNet

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

11 Commits
 
 
 
 
 
 
 
 

Repository files navigation

OBGNet

Code for ACM MM 2021 paper "Occlusion-aware Bi-directional Guided Network for Light Field Salient Object Detection"

Overall Architecture

Image text

Datasets

We train and evaluate our model on DUTLF-v2 dataset.

Requirements

  • Ubuntu 18.04
  • torch 1.7.1
  • python 3.8
  • opencv-python 4.5.3.56
  • imageio 2.4.1

Train

We train our model on multiple GPUs.

If you want to retrain our model, the process is as follows:

  1. Please adjust hyperparameters in 'train_multiGPUs.py' , such as 'root', 'batch_size', and so on.

  2. Please set CUDA devices in train_start.sh.

  3. cd to the code path, then start training.

    nohup sh train_start.sh > log/xxx.txt 2>&1 &

Please note that the 'batch_size' should be greater than or equal to 16 for model generalization.

Test

You can download the checkpoint we provided (Baidu Pan, code:uk24).

Adjust paths in 'test.py' and run it to obtain predictions.

We also provide results of our model (Baidu Pan, code:dy78).

Contact us

If you have any questions, please contact us (20120370@bjtu.edu.cn).

About

code for ACM MM 2021 paper "Occlusion-aware Bi-directional Guided Network for Light Field Salient Object Detection"

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published