Research map / Label-efficient and robust learning / Recognition and long-tail
Taxonomy-Aware Representation Alignment for Hierarchical Visual Recognition with Large Multimodal Models
Hulingxiao He, Zhi Tan, Yuxin Peng
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add TARA alignment to No-Thinking RFT | Qwen3-VL-2B · iNat21-Plant | HCA: 9.23 -> 12.78 | Improved |
| add TARA alignment to No-Thinking RFT | Qwen2.5-VL-3B · iNat21-Animal | HCA: 21.99 -> 24.02 | Improved |
| add TARA alignment on novel species | Qwen3-VL-2B · TerraIncognita novel | Order F1: 23.30 -> 33.45 | Improved |
| add TARA alignment on novel species at family level | Qwen3-VL-2B · TerraIncognita novel | Family F1: 11.47 -> 12.67 | No clear effect |
| add label alignment LC (first answer token) on top of visual alignment LV (ablation) | TARA · iNat21-Plant | HCA: 10.72 -> 12.78 | Improved |
| add LC alignment on top of LV (ablation) | TARA · iNat21-Plant | leaf accuracy: 33.27 -> 32.66 | No clear effect |
| move LC alignment to layer 28 from layer 14 (with LV at 14) (ablation) | TARA · iNat21-Plant | HCA: 10.40 -> 12.78 | Improved |
| use all visual tokens instead of last visual token for LV (ablation) | TARA · iNat21-Plant | HCA: 10.30 -> 12.78 | Improved |
| use first answer token instead of last question or mean answer tokens for LC (ablation) | TARA · iNat21-Plant | HCA: 12.01 -> 12.78 (last question token), 11.59 -> 12.78 (mean answer) | Improved |
| add TARA alignment to No-Thinking RFT | Qwen3-VL-2B · ImageWikiQA | accuracy: 48.70 -> 51.40 | Improved |
Closest papers by meaning
- Deep Taxonomic Networks for Unsupervised Hierarchical Prototype Discovery (NeurIPS 2025 · similarity 0.87)
- DynAlign: Unsupervised Dynamic Taxonomy Alignment for Cross-Domain Segmentation (ICLR 2025 · similarity 0.87)
- Topology-Aware Representation Alignment for Semi-Supervised Vision-Language Learning (NeurIPS 2026 · similarity 0.87)
- Taxonomy-Aware Evaluation of Vision-Language Models (CVPR 2025 · similarity 0.86)
- Democratizing Fine-grained Visual Recognition with Large Language Models (ICLR 2024 · similarity 0.86)
- Contrastive Visual Data Augmentation (ICML 2025 · similarity 0.86)
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.
