AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Multi-Label Cluster Discrimination for Visual Representation Learning

Xiang An, Kaicheng Yang, Xiangzi Dai, Ziyong Feng, Jiankang Deng

ECCV 2024 · Label-efficient and robust learning › Recognition and long-tail · approach: Partial and multi-label learning · 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
replace instance discrimination (OpenCLIP) with multi-label cluster discriminationViT-L/14 LAION-400M · ImageNet finetuneaccuracy: 86.2 -> 87.1Improved
assign top-8 positive centers per image instead of a single label (ablation)MLCD ViT-B/32 · ImageNet linear probeaccuracy: 71.4 -> 75.2Improved
increase positive centers from 8 to 32 (ablation)MLCD ViT-B/32 · ImageNet linear probeaccuracy: 75.2 -> 68.7Hurt
use adaptive similarity-threshold labels instead of fixed top-l (ablation)MLCD ViT-B/32 · ImageNet linear probeaccuracy: 75.2 -> 73.3Hurt
add separate positive and negative loss terms (MLCD vs MLC) (ablation)MLC loss · ImageNet linear probeaccuracy: 76.9 -> 78.1Improved
sample 10% of negative class centers instead of all (ablation)MLCD ViT-B/32 · ImageNet linear probeaccuracy: 63.2 -> 75.2Improved
sample only 1% of negative centers (ablation)MLCD ViT-B/32 · ImageNet linear probeaccuracy: 75.2 -> 73.4Hurt
increase number of classes from 1M to 5M (ablation)MLCD ViT-B/32 · ImageNet linear probeaccuracy: 75.2 -> 74.7Hurt
increase number of classes from 100K to 1M (ablation)MLCD ViT-B/32 · ImageNet linear probeaccuracy: 66.9 -> 75.2Improved
replace single-label UNICOM with MLCD on LAION-400MViT-B/32 · 26-dataset linear probe averageaverage accuracy: 79.2 -> 81.4Improved
replace CLIP vision tower with MLCD in LLaVA-style VLMQwen2-7B VLM · TextVQAaccuracy: 48.98 -> 42.52Hurt
replace CLIP vision tower with MLCD in LLaVA-style VLMQwen2-72B VLM · MathVista miniaccuracy: 38.3 -> 46.5Improved

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