Skip to content

Latest commit

 

History

History
92 lines (75 loc) · 3.28 KB

README.md

File metadata and controls

92 lines (75 loc) · 3.28 KB
Raw

RGB-D Salient Object Detection with Cross-Modality Modulation and Selection

Arxiv version: https://arxiv.org/abs/2007.07051

TensorFlow

TensorFlow implementation of cmMS for RGBD salient object detection

Requirements

Python 3
TensorFlow 1.x

Folder structure

The following shows the basic folder structure.

├── checkpoint
│   ├── coarse_224 # A pre-trained checkpoint (coarse.model-2850)
│   │   ├── checkpoint 
│   │   ├── coarse.model-2850.data-00000-of-00001 
│   │   ├── coarse.model-2850.index 
│   │   ├── coarse.model-2850.meta
├── vgg_pretrained # vgg pretrained model 
│   ├── imagenet-vgg-verydeep-16
│   ├── imagenet-vgg-verydeep-19
│   ├── vgg16.npy
│   ├── vgg19.npy
├── main_test.py # testing code
├── model.py # network
├── test_real. # put RGB images here
├── depth_real. # put depth images here
├── ops.py.py
├── utils.py
├── vgg.py

results

We rename the images, so the name of our result is different from the original data. For your evaluations, we also provide the corresponding renamed GT.

Google Drive: https://drive.google.com/file/d/1uu6Y_IDH6ukdkBkGN9zfYv4K-IYLc9aa/view?usp=sharing
Baidu Cloud:  https://pan.baidu.com/s/1eXmx0Tm3K5rEn7OlyPDLOg Password: 1234

Preporcessing

  1. download the pretrained VGG model
Google Drive: https://drive.google.com/file/d/1IDzr2OqoQk2LdecWRoReGnoIsPYYSL-J/view?usp=sharing
Baidu Cloud:  https://pan.baidu.com/s/1obO2IWLlkfVdXDj7gT5ogg Password: 1234
  1. download the checkpoint of our model, unzip, and put it to 'checkpoint/coarse_224' folder
Google Drive: https://drive.google.com/file/d/1YsQ4XBe1J3cho7BDM2hW85PpHX5m3-Yx/view?usp=sharing

Baidu Cloud:  https://pan.baidu.com/s/1txRl_-xNctC6x3mZAwyRbQ  Password: 1234

  1. normalize the depth maps (note that the foreground should have higher value than the background in our method) input=(input-min(min(input)))/(max(max(input))-min(min(input))) The step is very important for accurate results.

  2. resize the testing data to the size of 224*224 first normalize depth then resize will be better than first resize depth then normalize in our method. So please strictly follow our steps to generate testing data.

  3. put your rgb images to 'test_real' folder and your depth maps to 'depth_real' folder (paired rgb image and depth map should have same name)

Test

python main_test.py

find the results in the 'test_real' folder with the same name as the input image + "_out".

You can use a script to resize the results back to the same size as the original RGB-D image, or just use the results with a size of 224*224 for evaluations. We did not find much differences for the evaluation results.

Bibtex

@inproceedings{cmMS,
 author = {Li, Chongyi and Cong, Runmin and Piao Yongri and Xu Qianqian, and Loy, Chen Change},
 title = {RGB-D salient object eetection with cross-modality modulation and selection},
 booktitle = {Proceedings of the European Conference on Computer Vision (ECCV)},
 pages    = {},
 month = {},
 year = {2020}
}

Contact

If you have any questions, please contact Chongyi Li at lichongyi25@gmail.com.

Our recent related work for RGB-D Salient Object Detection

https://github.com/Li-Chongyi/ASIF-Net