Awesome

MoLF-NeurIPS19

Code repository for our paper entilted "Memory-oriented Decoder for Light Field Salient Object Detection".

• [Code] You can found in this link and the code is t551. Moreover, the checkpoint is here and the code is tjqy.
• [poster] A brief explanation for paper
• [Dataset] Our light field saliency datset(DUTLF-FS) is part of DUTLF. Our DUTLF-FS mainly rely on focal stacks in light field.
• [DUTLF-V2] DUTLF-V2 is the updated version of DUTLF-V1, which consists of 4204 light field samples. We are working on it and will make it publicly available soon.

Datasets

• Dataset: DUTLF

1. This dataset consists of DUTLF-MV, DUTLF-FS, DUTLF-Depth.
2. The dataset will be expanded to 3000 about real scenes.
3. We are working on it and will make it publicly available soon.

• Dataset: DUTLF-FS

1. DUTLF-FS is part of DUTLF dataset captured by Lytro camera, yet our dataset mainly rely on focal stack in light field for accurate salient object detection.
2. A large scale light field saliency dataset(DUTLF-FS) with 1465 paired images contains focal stack and RGB images. In addition, the dataset download link is here and the fetch code is vecy.
3. Three problematic images were deleted for better use in DUTLF-FS, and a detail description can be found in dataset download files.

Results

| DUTLF-FS | | LFSD | | HFUT |

• Note: Several samples can't synthesize 12 focal slices on LFSD, and so LFSD only contains 93 samples.

Citation

@inproceedings{Zhang_2019_MoLF,
  title={Memory-oriented Decoder for Light Field Salient Object Detection},
  author={Zhang, Miao and Li, Jingjing and Ji, Wei and Piao, Yongri and Lu, Huchuan},
  booktitle={Advances in Neural Information Processing Systems},
  pages={896--906},
  year={2019}
}

Contact Us

If you have any questions, please contact us ( weiji.dlut@gmail.com ).