Research map / Vision-language and multimodal / CLIP and zero-shot
Adaptive Token Refinement in Long-Tailed Large Vision-Language Models Fine-Tuning
Wenjun Miao, Mingda Li, Yanchao Hao, Zheng Wei
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 ATR instead of direct SFT | Qwen2.5-VL-7B fine-tuning · ImageNet-LT | overall accuracy: 78.2 -> 84.1 | Improved |
| add ATR instead of direct SFT | Qwen2.5-VL-7B fine-tuning · Places-LT | overall accuracy: 43.8 -> 51.4 | Improved |
| add ATR instead of prior-based reweighting | Qwen2.5-VL-7B fine-tuning · ImageNet-LT | overall accuracy: 79.3 -> 84.1 | Improved |
| add DFHT only (ablation) | ATR · ImageNet-LT | overall accuracy: 78.2 -> 81.8 | Improved |
| add ARLT only (ablation) | ATR · ImageNet-LT | overall accuracy: 78.2 -> 81.3 | Improved |
| add VTM only (ablation) | ATR · Places-LT | tail accuracy: 33.0 -> 39.5 | Improved |
| add VTM on top of DFHT and ARLT (ablation) | ATR · ImageNet-LT | overall accuracy: 82.0 -> 84.1 | Improved |
| add VTM on top of DFHT and ARLT (ablation) | ATR · Places-LT | overall accuracy: 49.6 -> 51.4 | Improved |
| use ATR instead of FisherSFT with DoRA | ATR · IconQA (Qwen2.5-VL, DoRA) | Avg accuracy: 74.0 -> 76.7 | Improved |
| use ATR instead of SFT on LLaVA-1.5-7B LoRA | ATR · IconQA | Avg accuracy: 68.2 -> 70.9 | Improved |
| lower filtering threshold tau from 0.8 to 0.7 (ablation) | ATR · ImageNet-LT | overall accuracy: 84.1 -> 83.6 | Hurt |
| raise filtering threshold tau from 0.8 to 0.9 (ablation) | ATR · ImageNet-LT | overall accuracy: 84.1 -> 83.8 | Hurt |
Closest papers by meaning
- From Head to Tail: Towards Balanced Representation in Large Vision-Language Models through Adaptive Data Calibration (CVPR 2025 · similarity 0.92)
- Fine-Grained Post-Training Quantization for Large Vision Language Models with Quantization-Aware Integrated Gradients (CVPR 2026 · similarity 0.89)
- One Layer's Trash is Another Layer's Treasure: Adaptive Layer-wise Visual Token Selection in LVLMs (CVPR 2026 · similarity 0.88)
- ATP-LLaVA: Adaptive Token Pruning for Large Vision Language Models (CVPR 2025 · similarity 0.88)
- Data Selection for Fine-tuning Vision Language Models via Cross Modal Alignment Trajectories (ICML 2026 · similarity 0.88)
- Language Bias in LVLMs: From In-Depth Analysis to Simple and Effective Mitigation (ICML 2026 · 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.
