RGB-D Salient Object Detection: A Survey

Authors: Tao Zhou, Deng-Ping Fan, Ming-Ming Cheng, Jianbing Shen, Ling Shao.

This is a survey to review related RGB-D SOD models along with benchmark datasets, and provide a comprehensive evaluation for these models. We also collect related review papers for SOD and light field SOD models. If you have papers to recommend or any suggestions, please feel free to contact us.

🏃 🏃 🏃 KEEP UPDATING.

Fig.0: A brief chronology of RGB-D based SOD. The first early RGB-D based SOD work was the DM model, proposed in 2012. Deep learning techniques have been widely applied to RGB-D based SOD since 2017. More details can be found in our paper.

Content:

1. Related Reviews and Surveys to SOD
2. RGB-D SOD Models
3. RGB-D SOD Datasets
4. Light Field SOD
   1. LF SOD Models
   2. LF Datasets
5. Evaluation
   1. Overall Evaluation
   2. Attribute-based Evaluation
6. RGB-D SOD Benchmark
7. Citation

Related Reviews and Surveys to SOD:

RGB-D SOD Models:

🔥🔥🔥Update (in 2022-02-16)

RGB-D SOD Datasets:

Light Field SOD:

LF SOD Models:

LF Datasets:

Evaluation:

Overall Evaluation:

Fig.1: A comprehensive evaluation for 24 representative RGB-D based SOD models, including LHM, ACSD, DESM, GP, LBE, DCMC, SE, CDCP, CDB, DF, PCF, CTMF, CPFP, TANet, AFNet, MMCI, DMRA, D3Net, SSF, A2dele, S2MA, ICNet, JL-DCF, and UC-Net. We obtain the terms of $S_{\alpha}$ and MAE values for the 24 models on five datasets (i.e., STERE, NLPR, LFSD, DES, and SIP ). We report the mean values of $S_{\alpha}$ and MAE across the five datasets. Note that these better models are shown in the upper left corner (\ie, with a larger $S_{\alpha}$ and smaller MAE).

======================= run evaluation code ===============================

1. We have computed values of different evaluation metrics for each image of each models and save as '***.mat', and the results can be downloaded from Google Drive or Baidu Drive(extraction code: urra).
2. Please unzip the downloaded file 'Sal_Det_Results_24_Models.zip' and put it into the file 'results';
3. To run 'run_overall_evaluation.m' (plot Fig.1 )

======================================================================

Fig.2: PR curves for 24 RGB-D based models on STERE, NLPR, LFSD, DES, SIP, GIT, SSD, and NJUD datasets.

Fig.3: F-measures under different thresholds for 24 RGB-D based models on STERE, NLPR, LFSD, DES, SIP, GIT, SSD, and NJUD datasets.

======================= run plot code ===========================

1. To run 'run_plot_curves.m' (plot Fig.2 and Fig.3)

=============================================================

Attribute-based Evaluation:

1. Downloading attribute datasets Baidu Drive (Code: tktz) or Google Drive.

RGB-D SOD Benchmark :

The complete RGB-D SOD benchmark can be found in this page: http://dpfan.net/d3netbenchmark/

Citation:

If you find our survey paper and evaluation code are useful, please cite the following paper:

@article{zhou2020rgbd,
	title={RGB-D Salient Object Detection: A Survey},
	author={Zhou, Tao and Fan, Deng-Ping and Cheng, Ming-Ming and Shen, Jianbing and Shao, Ling},
	journal={Computational Visual Media},
	pages={1--33},
	year={2021},
	publisher={Springer}
}