Cross-Image Relational Knowledge Distillation for Semantic Segmentation

This repository contains the source code of CIRKD (Cross-Image Relational Knowledge Distillation for Semantic Segmentation).

Requirement

Ubuntu 18.04 LTS

Python 3.8 (Anaconda is recommended)

CUDA 11.1

PyTorch 1.8.0

NCCL for CUDA 11.1

Backbones pretrained on ImageNet:

Performance on Cityscapes

All models are trained over 8 * NVIDIA GeForce RTX 3090

If you want to use mixed precision training, please add --fp16, see fp16 demo demo1, demo2, demo3.

Role	Network	Method	Val mIoU	test mIoU	Pretrained	train script
Teacher	DeepLabV3-ResNet101	-	78.07	77.46	Google Drive	sh
Student	DeepLabV3-ResNet18	Baseline	74.21	73.45	-	sh
Student	DeepLabV3-ResNet18	CIRKD	76.38	75.05	Google Drive	sh
Student	DeepLabV3-MobileNetV2	Baseline	73.12	72.36	-	sh
Student	DeepLabV3-MobileNetV2	CIRKD	75.42	74.03	Google Drive	sh
Student	PSPNet-ResNet18	Baseline	72.55	72.29	-	sh
Student	PSPNet-ResNet18	CIRKD	74.73	74.05	Google Drive	sh

Performance of Segmentation KD methods on Cityscapes

Method	Val mIoU	Val mIoU	train script
Teacher	DeepLabV3-ResNet101	DeepLabV3-ResNet101
Baseline	78.07	78.07
Student	DeepLabV3-ResNet18	DeepLabV3-MobileNetV2
Baseline	74.21	73.12
SKD [3]	75.42	73.82	sh
IFVD [4]	75.59	73.50	sh
CWD [5]	75.55	74.66	sh
DSD [6]	74.81	74.11	sh
CIRKD [7]	76.38	75.42

The references are shown in references.md

Evaluate pre-trained models on Cityscapes test sets

You can run test_cityscapes.sh. You can zip the resulting images and submit it to the Cityscapes test server.

Note: The current codes have been reorganized and we have not tested them thoroughly. If you have any questions, please contact us without hesitation.

Performance of Segmentation KD methods on Pascal VOC

The Pascal VOC dataset for segmentation is available at Baidu Drive

Role	Network	Method	Val mIoU	train script
Teacher	DeepLabV3-ResNet101	-	77.67	sh
Student	DeepLabV3-ResNet18	Baseline	73.21	sh
Student	DeepLabV3-ResNet18	CIRKD	74.50	sh
Student	PSPNet-ResNet18	Baseline	73.33	sh
Student	PSPNet-ResNet18	CIRKD	74.78	sh

Performance of Segmentation KD methods on CamVid

The CamVid dataset for segmentation is available at Baidu Drive

Role	Network	Method	Val mIoU	train script
Teacher	DeepLabV3-ResNet101	-	69.84	sh
Student	DeepLabV3-ResNet18	Baseline	66.92	sh
Student	DeepLabV3-ResNet18	CIRKD	68.21	sh
Student	PSPNet-ResNet18	Baseline	66.73	sh
Student	PSPNet-ResNet18	CIRKD	68.65	sh

Citation

@inproceedings{yang2022cross,
  title={Cross-image relational knowledge distillation for semantic segmentation},
  author={Yang, Chuanguang and Zhou, Helong and An, Zhulin and Jiang, Xue and Xu, Yongjun and Zhang, Qian},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={12319--12328},
  year={2022}
}

Muyun99/CIRKD