AI Paper Atlas

Research map / Detection and segmentation / Open-set and few-shot detection

Unsupervised Object-Level Representation Learning from Scene Images

Jiahao Xie, Xiaohang Zhan, Ziwei Liu, Yew-Soon Ong, Chen Change Loy

NeurIPS 2021 · Detection and segmentation › Open-set and few-shot detection · approach: Object-centric learning theory and video · 7 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 intra-RoI loss (ablation)ORL · VOC07mAP: 84.5 -> 85.7Improved
add inter-RoI loss (ablation)ORL · VOC07mAP: 84.5 -> 85.9Improved
add both intra-RoI and inter-RoI losses (ablation)ORL · VOC07mAP: 84.5 -> 86.7Improved
increase nearest neighbors K from 1 to 10 (ablation)ORL · VOC07mAP: 84.7 -> 86.7Improved
increase nearest neighbors K from 10 to 20 (ablation)ORL · VOC07mAP: 86.7 -> 87.0Improved
vary RoI pairs N from 5% to 10% to 20% (ablation)ORL · VOC07mAP: 86.1 -> 86.7 -> 86.4No clear effect
add four extra 96x96 crops to BYOL (multi-crop) (ablation)BYOL · VOC07mAP: 84.5 -> 84.0Hurt
object-level ORL vs multi-crop BYOL with same views (ablation)ORL · VOC07mAP: 84.0 -> 86.7Improved
replace selective search boxes with COCO ground-truth boxes (ablation)ORL · VOC07mAP: 86.7 -> 85.4Hurt
pre-training schedule 100 to 1600 epochs (ablation)ORL vs BYOL · VOC07mAP: 83.5 -> 87.1Improved
ORL at 200 epochs vs BYOL at 1600 epochs (ablation)ORL · VOC07mAP: 84.9 -> 85.2Improved
ORL vs BYOL pre-trained on COCO (ablation)ORL · Places205 lineartop-1: 50.5 -> 52.7Improved

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