Code for ACM MM 2021 paper "Occlusion-aware Bi-directional Guided Network for Light Field Salient Object Detection"
We train and evaluate our model on DUTLF-v2 dataset.
- Ubuntu 18.04
- torch 1.7.1
- python 3.8
- opencv-python 4.5.3.56
- imageio 2.4.1
We train our model on multiple GPUs.
If you want to retrain our model, the process is as follows:
-
Please adjust hyperparameters in 'train_multiGPUs.py' , such as 'root', 'batch_size', and so on.
-
Please set CUDA devices in train_start.sh.
-
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.
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).
If you have any questions, please contact us (20120370@bjtu.edu.cn).