AI Paper Atlas

Research map / Continual learning / Class-incremental

Early Preparation Pays Off: New Classifier Pre-tuning for Class Incremental Semantic Segmentation

Zhengyuan Xie, Haiquan Lu, Jia-wen Xiao, Enguang Wang, Le Zhang, Xialei Liu

ECCV 2024 · Continual learning › Class-incremental · approach: Unknown-class mining for incremental segmentation · 18 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add NeST new-classifier pre-tuning to MiBMiB · Pascal VOC 2012 15-1 (ResNet-101)mIoU (all): 38.2 -> 51.8Improved
add NeST to MiB in the hard 10-1 settingMiB · Pascal VOC 2012 10-1 (ResNet-101)mIoU (all): 10.2 -> 37.4Improved
add NeST to MiB with Swin-BMiB · Pascal VOC 2012 15-1 (Swin-B)mIoU (all): 36.9 -> 71.4Improved
add NeST to PLOPPLOP · Pascal VOC 2012 15-1 (ResNet-101)mIoU (all): 56.2 -> 63.1Improved
add NeST to RCILRCIL · Pascal VOC 2012 15-1 (ResNet-101)mIoU (all): 58.9 -> 61.4Improved
add NeST to MiB in the 19-1 settingMiB · Pascal VOC 2012 19-1 (ResNet-101)mIoU (all): 69.6 -> 69.7No clear effect
add NeST to MiB on ADE20K 100-5MiB · ADE20K 100-5 (ResNet-101)mIoU (all): 25.9 -> 32.7Improved
add NeST to MiB on ADE20K 100-50MiB · ADE20K 100-50 (ResNet-101)mIoU (all): 34.9 -> 35.1No clear effect
use NeST transformation init instead of random classifier init (ablation)MiB · Pascal VOC 2012 15-1mIoU (all): 34.1 -> 51.8Improved
tune background-initialized classifier directly (two-stage) vs NeST (ablation)MiB · Pascal VOC 2012 15-1mIoU (all): 38.7 -> 51.8Improved
use designed initialization of importance and projection matrices instead of random (ablation)NeST · Pascal VOC 2012 15-1mIoU (new classes 16-20): 7.5 -> 20.4Improved
add importance matrix on top of projection matrix (ablation)NeST · Pascal VOC 2012 15-1mIoU (all): 44.3 -> 51.8Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy