Research map / Vision-language and multimodal / CLIP and zero-shot
HyFL-CLIP: Hyperbolic Fine-Tuning of CLIP for Robust Long-Context Understanding
jiha jang, Hayeon Kim, Junghun James Kim, Chulwon Lee, Se Young Chun
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 entailment loss L_ent (ablation) | HyFL-CLIP · Average of all retrieval benchmarks | retrieval accuracy: 69.3 -> 69.8 | Improved |
| add cross-manifold distillation L_distill (ablation) | HyFL-CLIP · Average of all retrieval benchmarks | retrieval accuracy: 68.3 -> 69.3 | Improved |
| add L_distill (short-caption retrieval) (ablation) | HyFL-CLIP ViT-L/14 · COCO I2T | Top-1: 42.77 -> 41.56 | Improved |
| add L_ent to improve perturbation robustness (ablation) | HyFL-CLIP · Long-caption retrieval under perturbation | performance degradation (%): 35.64 -> 32.38 | Improved |
| add L_distill for perturbation robustness (ablation) | HyFL-CLIP · Long-caption retrieval under perturbation | performance degradation (%): 32.28 -> 32.38 | No clear effect |
| use hyperbolic distance instead of Euclidean for ranking perturbed captions | HyFL-CLIP vs HiMo-CLIP · Perturbed caption ranking | mean rank: 184.94 -> 156.99 | Improved |
| rank by hyperbolic entailment instead of hyperbolic distance (ablation) | HyFL-CLIP · Perturbed caption ranking | mean rank: 156.99 -> 78.50 | Improved |
| hyperbolic fine-tuning vs HiMo-CLIP under word dropout | HyFL-CLIP · Long-caption retrieval, word dropout p=0.5 | Top-1 accuracy: 58.75 -> 70.20 | Improved |
| hyperbolic fine-tuning vs HiMo-CLIP on Urban-1k ViT-B/16 | HyFL-CLIP · Urban-1k | I2T Top-1: 89.20 -> 91.80 | Improved |
| hyperbolic fine-tuning vs HiMo-CLIP on DOCCI ViT-L/14 I2T | HyFL-CLIP · DOCCI | I2T Top-1: 82.35 -> 82.12 | No clear effect |
| vary lambda1 from 0.01 to 0.20 (ablation) | HyFL-CLIP · Retrieval (I2T/T2I) | accuracy: 74.58/75.35 -> 75.35/75.75 | No clear effect |
| hyperbolic fine-tuning vs HiMo-CLIP on short COCO I2T | HyFL-CLIP · COCO | I2T: 60.84 -> 58.20 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- HiMo-CLIP — HiMo-CLIP: Modeling Semantic Hierarchy and Monotonicity in Vision-Language Alignment (AAAI 2026) · better on 17 of 17 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Retaining Knowledge and Enhancing Long-Text Representations in CLIP through Dual-Teacher Distillation (CVPR 2025 · similarity 0.90)
- GOAL: Global-local Object Alignment Learning (CVPR 2025 · similarity 0.90)
- Understanding Fine-tuning CLIP for Open-vocabulary Semantic Segmentation in Hyperbolic Space (CVPR 2025 · similarity 0.89)
- SmartCLIP: Modular Vision-language Alignment with Identification Guarantees (CVPR 2025 · similarity 0.89)
- FineLIP: Extending CLIP's Reach via Fine-Grained Alignment with Longer Text Inputs (CVPR 2025 · similarity 0.89)
- FG-CLIP: Fine-Grained Visual and Textual Alignment (ICML 2025 · similarity 0.89)
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.
