RD3D: RGB-D Salient Object Detection via 3D Convolutional Neural Networks

This repo is the official implementation of "RD3D: RGB-D Salient Object Detection via 3D Convolutional Neural Networks" by Qian Chen, Ze Liu, Yi Zhang, Keren Fu, Qijun Zhao and Hongwei Du. pdf: https://arxiv.org/abs/2101.10241

Main Results

Dataset	S_α	F_β^max	E_Φ^max	MAE
NJU2K	0.916	0.914	0.947	0.036
NLPR	0.930	0.919	0.965	0.022
STERE	0.911	0.906	0.947	0.037
RGBD135	0.935	0.929	0.972	0.019
DUTLF-D	0.932	0.939	0.960	0.031
SIP	0.885	0.889	0.924	0.048

All results can be found in:

BaiDu: https://pan.baidu.com/s/132ChkfOY9hDQ4FO4ada2Mg, password: 3e96.

Installation

Requirements

Ubuntu 16.04
python=3.6
pytorch>=1.3
torchvision with pillow<7
cuda>=10.1
others: pip install termcolor opencv-python tensorboard

Datasets

Our training data can be download from:

BaiDu: https://pan.baidu.com/s/1uF6LxbH0RIcMFN71cEcGHQ, password: 5z48

or:

Google Drive: https://drive.google.com/open?id=1BpVabSlPH_GhozzRQYjxTOT_cS6xDUgf

Usage

Training

python train.py --data_dir <data directory> [--output_dir <output directory>]

Evaluation

python test.py --data_dir <data directory> --model_path <model directory> --multi_load

We follow the RGB-D SOD benchmark setting from: http://dpfan.net/d3netbenchmark/

Citation

@article{chen2021rd3d,
  title={RGB-D Salient Object Detection via 3D Convolutional Neural},
  author={Chen, Qian and Liu,Ze and Zhang, Yi and Fu, Keren and Zhao, Qijun and Du, Hongwei},
  journal={AAAI},
  year={2021}
}

License

The code is released under MIT License (see LICENSE file for details).

sunpeng1996/RD3D