Skip to content

Latest commit

 

History

History
87 lines (50 loc) · 6 KB

README.md

File metadata and controls

87 lines (50 loc) · 6 KB
Raw

Leveraging Instance-, Image- and Dataset-Level Information for Weakly Supervised Instance Segmentation

Code contact e-mail: Yu-Huan Wu (wuyuhuan (at) mail(dot)nankai(dot)edu(dot)cn)

Introduction

Weakly supervised semantic instance segmentation with only image-level supervision, instead of relying on expensive pixel-wise masks or bounding box annotations, is an important problem to alleviate the data-hungry nature of deep learning. In this paper, we tackle this challenging problem by aggregating the image-level information of all training images into a large knowledge graph and exploiting semantic relationships from this graph. Specifically, our effort starts with some generic segment-based object proposals (SOP) without category priors. We propose a multiple instance learning (MIL) framework, which can be trained in an end-to-end manner using training images with image-level labels. For each proposal, this MIL framework can simultaneously compute probability distributions and category-aware semantic features, with which we can formulate a large undirected graph. The category of background is also included in this graph to remove the massive noisy object proposals. An optimal multi-way cut of this graph can thus assign a reliable category label to each proposal. The denoised SOP with assigned category labels can be viewed as pseudo instance segmentation of training images, which are used to train fully supervised models. The proposed approach achieves state-of-the-art performance for both weakly supervised instance segmentation and semantic segmentation.

Citation

If you are using the code/model/data provided here in a publication, please consider citing:

@article{liu2020leveraging,
  title={Leveraging Instance-, Image- and Dataset-Level Information for Weakly Supervised Instance Segmentation},
  author={Yun Liu and Yu-Huan Wu and Peisong Wen and Yujun Shi and Yu Qiu and Ming-Ming Cheng},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
  volume={44},
  number={3},
  pages={1415--1428},
  year={2020},
  doi={10.1109/TPAMI.2020.3023152},
  publisher={IEEE}
}

Requirements

  • Python 3.5, PyTorch 0.4.1, Torchvision 0.2.2.post3, CUDA 9.0
  • Validated on Ubuntu 16.04, NVIDIA TITAN Xp

Testing LIID

  1. Clone the LIID repository

    git clone https://github.com/yun-liu/LIID.git

  2. Download the pretrained model of the MIL framework, and put them into $ROOT_DIR folder.

  3. Download the Pascal VOC2012 dataset. Extract the dataset files into $VOC2012_ROOT folder.

  4. Download the segment-based object proposals, and extract the data into $VOC2012_ROOT/proposals/ folder.

  5. Download the compiled binary files, and put the binary files into $ROOT_DIR/cut/multiway_cut/.

  6. Change the path in cut/run.sh to your own project root.

  7. run ./make.sh to build CUDA dependences.

  8. Run python3 gen_proposals.py. Remember to change the voc-root to your own $VOC2012_ROOT. The proposals with labels will be generated in the $ROOT_DIR/proposals folder.

Pretrained Models and data

The pretrained model of the MIL framework can be downloaded here.

The Pascal VOC2012 dataset can be downloaded here or other mirror websites.

S4Net proposals used for testing can be downloaded here.

The 24K simple ImageNet data (including S4Net proposals) can be downloaded here.

MCG proposals can be downloaded here.

Training with Pseudo Labels

For instance segmentation, you can use official or popular public Mask R-CNN projects like mmdetecion, Detectron2, maskrcnn-benchmark, or other popular open-source projects.

For semantic segmentation, you can use official Caffe implementation of deeplab, third-party PyTorch implementation here, or third-party Tensorflow Implementation here.

Precomputed Results

Results of instance segmentation on the Pascal VOC2012 segmentation val split can be downloaded here.

Results of semantic segmentation trained with 10K images, 10K images + 24K simple ImageNet images, 10K images (Res2Net-101) on the Pascal VOC2012 segmentation val split can be downloaded here.

Other Notes

Since it is difficult to install and configure IBM CPLEX, for convenience, we provide the compiled binary file which can run directly. If you desire to get the complete source code for solving the multi-way cut and ensure that there is no commercial use of it, please contact Yu-Huan Wu (wuyuhuan (at) mail(dot)nankai(dot)edu(dot)cn).

Acknowledgment

This code is based on IBM CPLEX. Thanks to the IBM CPLEX academic version.

License

The code is released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International Public License for NonCommercial use only. Any commercial use should get formal permission first (Email: VAGRANTLYUN@GMAIL.COM).