IRG for Knowledge Distillation

IRG is an open source implementation of the paper called "Knowledge Distillation via Instance Relationship Graph":

Yufan Liu, Jiajiong Cao, Bing Li, Chunfeng Yuan, Weiming Hu et al. Knowledge Distillation via Instance Relationship Graph. IEEE Conference on Computer Vision and Pattern Recognition (CVPR): Long Beach, USA. 2019.06.16-2019.06.20.

Abstract

The key challenge of knowledge distillation is to extract general, moderate and sufficient knowledge from a teacher network to guide a student network. In this paper, a novel Instance Relationship Graph (IRG) is proposed for knowledge distillation. It models three kinds of knowledge, including instance features, instance relationships and feature space transformation, while the latter two kinds of knowledge are neglected by previous methods. Firstly, the IRG is constructed to model the distilled knowledge of one network layer, by considering instance features and instance relationships as vertexes and edges respectively. Secondly, an IRG transformation is proposed to models the feature space transformation across layers. It is more moderate than directly mimicking the features at intermediate layers. Finally, hint loss functions are designed to force a student’s IRGs to mimic the structures of a teacher’s IRGs. The proposed method effectively captures the knowledge along the whole network via IRGs, and thus shows stable convergence and strong robustness to different network architectures. In addition, the proposed method shows superior performance over existing methods on datasets of various scales.

Citation

If you find IRG useful in your research, please cite our paper:

@inproceedings{liu2019knowledge,
  title={Knowledge Distillation via Instance Relationship Graph},
  author={Liu, Yufan and Cao, Jiajiong and Li, Bing and Yuan, Chunfeng and Hu, Weiming and Li, Yangxi and Duan, Yunqiang},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  pages={7096--7104},
  year={2019}
}

Framework

Requirements

Caffe-windows

Usage

Put *.cpp/*.hpp/*.cu files into corresponding directories of caffe.
Compile caffe.

  make all -j 8

Use prototxt files (e.g., IRG.prototxt) to train. IRG.prototxt is the train prototxt of single IRG (Teacher: ResNet20; Student: ResNet20-x0.5). IRG_Transformation.prototxt is the train prototxt of single IRG Transformation (Teacher: ResNet20; Student: ResNet20-x0.5). ResNet-20.prototxt is the original train prototxt of ResNet20.

Experiments

Contact

If any question, please contact yufan.liu@ia.ac.cn, or use public issues section of this repository.

yufanLIU/IRG