ilt

Introduction

Official Repo

Code Snippet

ILT (ICCVW'2019)

@inproceedings{michieli2019incremental,
  title={Incremental learning techniques for semantic segmentation},
  author={Michieli, Umberto and Zanuttigh, Pietro},
  booktitle={Proceedings of the IEEE/CVF international conference on computer vision workshops},
  pages={0--0},
  year={2019}
}

Results

PASCAL VOC

PASCAL VOC

Backbone	Crop Size	Setting	mIoU	Download
R-101-D16	512x512	15-5-disjoint		cfg | modellinks-per-step | log
R-101-D16	512x512	15-5-overlapped		cfg | modellinks-per-step | log
R-101-D16	512x512	15-1-disjoint	8.6%	cfg | modellinks-per-step | log
R-101-D16	512x512	15-1-overlapped	9.3%	cfg | modellinks-per-step | log
R-101-D16	512x512	10-1-disjoint	5.4%	cfg | modellinks-per-step | log
R-101-D16	512x512	10-1-overlapped	5.9%	cfg | modellinks-per-step | log

ADE20k

Backbone	Crop Size	Setting	mIoU	Download
R-101-D16	512x512	100-50-disjoint		cfg | modellinks-per-step | log
R-101-D16	512x512	100-50-overlapped		cfg | modellinks-per-step | log
R-101-D16	512x512	100-10-disjoint		cfg | modellinks-per-step | log
R-101-D16	512x512	100-10-overlapped		cfg | modellinks-per-step | log
R-101-D16	512x512	100-5-disjoint		cfg | modellinks-per-step | log
R-101-D16	512x512	100-5-overlapped		cfg | modellinks-per-step | log

More

You can also download the model weights from following sources:

• BaiduNetdisk: https://pan.baidu.com/s/1e9LlD6ITuLECstFcVPaCxQ with access code 55mq