Fine-R1 Delivers Few-Shot Fine-Grained Visual Recognition
Researchers post an arXiv preprint on Feb 7, 2026 introducing Fine-R1, a multimodal large language model tailored for fine-grained visual recognition using Chain-of-Thought supervised fine-tuning and Triplet Augmented Policy Optimization. With only 4-shot training, the model reportedly outperforms general MLLMs and contrastive CLIP models on seen and unseen sub-categories, improving robustness to intra-class variance and discriminative ability; code is available.
Key Points
- 1Introduces Fine-R1, an MLLM trained with Chain-of-Thought and triplet-augmented policy optimization.
- 2Improves discrimination by mixing intra-class trajectories and maximizing inter-class response distinctions.
- 3Enables 4-shot recognition of seen and unseen sub-categories, reducing annotation requirements for practitioners.
Scoring Rationale
Strong methodological novelty and few-shot results drive score; limited by single arXiv preprint without peer review.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
