Research map / Vision-language and multimodal / Human and embodied modeling
LOREAL: Mitigating Low-Resolution Challenges in Vision-Language Models with Attribute-driven Prompt Self-Distillation
Xucong Wang, Pengkun Wang, Zhe Zhao, Liheng Yu, Rui Mao, Yang Wang
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 LOREAL to MMA at 96x96 | MMA · LR-B2N average, 96x96 | HM: 40.50 -> 63.14 | Improved |
| add LOREAL to MMRL at 96x96 | MMRL · LR-B2N average, 96x96 | HM: 38.90 -> 61.71 | Improved |
| add LOREAL to CoOp at 96x96 | CoOp · LR-B2N average, 96x96 | HM: 35.54 -> 47.48 | Improved |
| add LOREAL near standard resolution (192x192) | MMA · LR-B2N average, 192x192 | HM: 78.27 -> 79.70 | Improved |
| add LOREAL to MMRL | MMRL · LR-CE target average, 96x96 | accuracy: 40.77 -> 55.08 | Improved |
| add LOREAL to MMRL | MMRL · LR-DG target average, 96x96 | accuracy: 15.42 -> 30.64 | Improved |
| add LOREAL, extra cost of meta-nets | MaPLe · 96x96 efficiency | tunable parameters / HM: +104.5K params, +4ms train, HM 34.85 -> 57.25 | Improved |
| add LR to standard-resolution student direction only vs both directions (ablation) | LOREAL · LR-B2N average, 144x144 | HM: 68.91 -> 69.78 | Improved |
| add LLD and HLD distillation losses (ablation) | LOREAL · LR-B2N average, 144x144 | HM: 69.78 -> 71.73 | Improved |
| low-level distillation is more important than high-level distillation (ablation) | LOREAL · LR-B2N average, 144x144 | HM | Improved |
| increase meta-net dimension Ds from 32 to 2048 (ablation) | LOREAL · LR-B2N average, 144x144 | HM: 71.73 -> 67.60 | Hurt |
| increase attribute prompt length M (ablation) | LOREAL · LR-B2N | HM: about -2% | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- MMRL — MMRL: Multi-Modal Representation Learning for Vision-Language Models (CVPR 2025) · better on 5 of 5 shared results
- MaPLe — MaPLe: Multi-Modal Prompt Learning (CVPR 2023) · better on 1 of 1 shared result
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
- PromptKD: Unsupervised Prompt Distillation for Vision-Language Models (CVPR 2024 · similarity 0.90)
- ArGue: Attribute-Guided Prompt Tuning for Vision-Language Models (CVPR 2024 · similarity 0.89)
- IPO: Interpretable Prompt Optimization for Vision-Language Models (NeurIPS 2024 · similarity 0.88)
- Learning to Prompt with Text Only Supervision for Vision-Language Models (AAAI 2025 · similarity 0.88)
- CoPL: Contextual Prompt Learning for Vision-Language Understanding (AAAI 2024 · similarity 0.88)
- Prompt Tuning In a Compact Attribute Space (AAAI 2025 · similarity 0.88)
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.
