Home

Awesome

MSRNet-CVIU

Instance-Level Salient Object Segmentation <br /> Guanbin Li, Pengxiang Yan, YuanXie, Guisheng Wang, Liang Lin, Yizhou Yu. <br /> Computer Vision and Image Understanding (CVIU), 2021, Elsevier. <br /> [Paper] <br />

<img src="figures/intro.png" height="125px">

Salient Object Detection

Links

MSRNet-MXNet

This code MSNet-MXNet is tested on Ubuntu 16.04, Python=3.7 (via Anaconda3), MXNet=1.3.1, CUDA=9.2.

Install

# install MXNet (refer to your CUDA version)
$ pip install mxnet-cu92==1.3.1
# install others
$ pip install mxboard pyyaml tqdm opencv-python Pillow

Clone this repository with submodules:

git clone --recurse-submodules https://github.com/Kinpzz/MSRNet-CVIU.git

Training

# Training on DUTS-TR for salient object/region detection
$ python train.py --config config/MSRNet_DUTS.yaml

Inference

Download and save model weights in MSRNet-MXNet/models.

# Inference on DUTS-TE for salient object/region detection
# modify the config file to inference on other datasets
$ python test.py --config config/MSRNet_DUTS.yaml

Salient Instance Segmentation

Links

Demo

Run instance_seg/demo.m in MATLAB. Note that the salient region maps & salient contour maps are predicted by the above-mentioned MSRNet-MXNet fine-tuned on ILSO datasets.

Citation

If you find this work helpful, please consider citing

# CVIU 2021
@article{li2021instance,
  title = {Instance-level salient object segmentation},
  author = {Li, Guanbin and Yan, Pengxiang and Xie, Yuan and Wang, Guisheng and Lin, Liang and Yu, Yizhou},
  journal = {Computer Vision and Image Understanding},
  volume = {207},
  pages = {103207},
  year = {2021},
  issn = {1077-3142},
}
# CVPR 2017
@inproceedings{li2017instance,
  title={Instance-level salient object segmentation},
  author={Li, Guanbin and Xie, Yuan and Lin, Liang and Yu, Yizhou},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  pages={2386--2395},
  year={2017}
}