A2dele: Adaptive and Attentive Depth Distiller for Efficient RGB-D Salient Object Detection

Introduction

Accepted paper in CVPR2020, 'A2dele: Adaptive and Attentive Depth Distiller for Efficient RGB-D Salient Object Detection', Yongri Piao, Zhengkun Rong, Miao Zhang, Weisong Ren and Huchuan Lu.

Usage Instructions

Requirements

  • Windows 10
  • PyTorch 0.4.1
  • CUDA 9.0
  • Cudnn 7.6.0
  • Python 3.6.5
  • Numpy 1.16.4

Training and Testing Datasets

Training dataset

Testing dataset

Depth Stream

Training

  • Modify your path of training dataset in train_depth
  • Run train_depth

Testing

  • Download pretrained depth model from here. Code: uklr
  • Modify your path of testing dataset in test_depth
  • Run test_depth to inference saliency maps
  • Saliency maps generated from the depth stream can be downnloaded from here. Code: 2e3l

RGB Stream

Training

  • Modify your path of training dataset in train_RGB
  • Modify the pretrained depth model path
  • Run train_RGB

Testing

  • Download pretrained RGB model from here. Code: tj7b
  • Modify your path of testing dataset in test_depth
  • Run test_RGB to inference saliency maps
  • Saliency maps generated from the RGB stream can be downnloaded from here. Code: tb3y

Contact and Questions

Contact: Zhengkun Rong. Email: 18642840242@163.com or rzk911113@mail.dlut.edu.cn