Research map / Vision-language and multimodal / Multimodal LLMs
Paying More Attention to Visual Tokens in Self-Evolving Large Multimodal Models
Shravan Venkatraman, Ritesh Thawkar, Omkar Thawakar, Rao M Anwer, Hisham Cholakkal, Salman Khan, Fahad Shahbaz Khan
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 geometric invariance reward only (ablation) | VISE on Qwen3-VL-2B · COCO Caption | CIDEr: 21.54 -> 26.37 | Improved |
| add semantic invariance (ghosting) reward only (ablation) | VISE on Qwen3-VL-2B · COCO Caption | CIDEr: 21.54 -> 35.53 | Improved |
| combine geometric reward with semantic reward (ablation) | VISE on Qwen3-VL-2B · COCO Caption | CIDEr: 35.53 -> 38.39 | Improved |
| add geometric reward to semantic reward (ablation) | VISE on Qwen3-VL-2B · COCO Cap Chair-I | Chair-I: 9.06 -> 8.21 | Improved |
| train with VISE on unlabeled COCO images | Qwen3-VL-2B · COCO Caption | CIDEr: 21.54 -> 38.39 | Improved |
| train with VISE | Qwen3-VL-2B · COCO Cap Chair-I | Chair-I: 13.21 -> 8.21 | Improved |
| train with VISE | Qwen3-VL-2B · POPE | accuracy: 89.01 -> 90.03 | Improved |
| train with self-consistency self-evolution (EvoLMM) | Qwen3-VL-2B · COCO Caption | CIDEr: 21.54 -> 20.84 | Hurt |
| train with EvoLMM | Qwen3-VL-2B · OK-VQA | accuracy: 40.76 -> 38.03 | Hurt |
| apply VISE to larger models | VISE · COCO Caption | CIDEr gain: +16.85 at 2B to +8.72 at 32B | Mixed |
| apply VISE to four backbone families | VISE · COCO Caption | CIDEr: Gemma3-12B 22.18 -> 29.83; Llama-3.2-11B 16.37 -> 22.81 | Improved |
| add semantic invariance on 8B model (ablation) | VISE on Qwen3-VL-8B · COCO Caption | CIDEr: 29.01 -> 35.27 (R_sem only) | Improved |
Closest papers by meaning
- VisPlay: Self-Evolving Vision-Language Models (CVPR 2026 · similarity 0.91)
- Vision-OPD: Learning to See Fine Details for Multimodal LLMs via On-Policy Self-Distillation (NeurIPS 2026 · similarity 0.90)
- Agent0-VL: Exploring Self-Evolving Agent for Tool-Integrated Vision-Language Reasoning (ICML 2026 · similarity 0.90)
- Diving into Self-Evolving Training for Multimodal Reasoning (ICML 2025 · similarity 0.89)
- Unveiling the Tapestry of Consistency in Large Vision-Language Models (NeurIPS 2024 · similarity 0.89)
- Learn to Think: Improving Multimodal Reasoning through Vision-Aware Self-Improvement Training (ICML 2026 · 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.
