AI Paper Atlas

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

Free-Grained Hierarchical Visual Recognition

Seulki Park, Zilin Wang, Stella X. Yu

CVPR 2026 · Label-efficient and robust learning › Recognition and long-tail · approach: Correspondence and causal representation learning

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
train hierarchical classifier with free-grain labels instead of full labelsH-CAST · CUB-FFull-path accuracy (%): 84.9 -> 45.1Hurt
train hierarchical classifier with free-grain labels instead of full labelsH-CAST · iNat21-mini-FFull-path accuracy (%): 64.9 -> 25.6Hurt
add text-guided pseudo attributes (contrastive image-text loss) to H-CASTH-CAST · ImageNet-FFull-path accuracy (%): 57.59 -> 63.20Improved
add text-guided pseudo attributes to H-CASTH-CAST · iNat21-mini-FFull-path accuracy (%): 25.63 -> 29.74Improved
add taxonomy-guided semi-supervised learning (Taxon-SSL) instead of H-CASTH-ViT · iNat21-mini-FFull-path accuracy (%): 25.63 -> 31.74Improved
use Taxon-SSL instead of H-CASTH-ViT · ImageNet-FFull-path accuracy (%): 57.59 -> 48.40Hurt
combine Taxon-SSL with Text-Attr (ablation)Taxon-SSL · ImageNet-FFull-path accuracy (%): 48.40 -> 49.65Improved
combine Taxon-SSL with Text-Attr (ablation)Taxon-SSL · iNat21-mini-FFull-path accuracy (%): 31.74 -> 31.93Improved
use Text-Attr rather than Taxon-SSL when fine-grained labels are scarceImageNet-F classes · ImageNet-F per-class fine accuracyper-class accuracyMixed
use consistency-based stopping instead of confidence-based stopping (tau=0.9) (ablation)Free-grained inference, Text-Attr (H-CAST) · ImageNet-FFine-level reached and correct (%): 48% -> 63%Improved
add text-guided pseudo attributes to H-CAST on ImageNet-FH-CAST · ImageNet-FTICE (%): 21.81 -> 18.58Improved

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