Research map / Vision-language and multimodal / Multimodal LLMs
Visual Reasoning: From State to Transformation
Xin Hong, Yanyan Lan, Liang Pang, Jiafeng Guo, Xueqi Cheng
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 REINFORCE with correctness reward (ablation) | ResNet- -T · TRANCE Event | Acc: 0.6553 -> 0.6729 | Improved |
| add REINFORCE with distance reward (ablation) | ResNet- -T · TRANCE Event | Acc: 0.6553 -> 0.6734 | Improved |
| add REINFORCE with correctness and distance rewards (ablation) | ResNet- -T · TRANCE Event | Acc: 0.6553 -> 0.6740 | Improved |
| use transformer decoder instead of GRU | ResNet- encoder · TRANCE Event | Acc: 0.6057 -> 0.6553 | Improved |
| use early-fusion ResNet encoder instead of late-fusion BCNN | GRU decoder · TRANCE Basic | Acc: 0.9524 -> 0.9814 | Improved |
| use larger ResNet encoder instead of 4-layer CNN | GRU decoder, subtraction fusion · TRANCE Event | Acc: 0.4419 -> 0.6057 | Improved |
| vary final-state camera view (Event to View) | ResNet- -T · TRANCE | Acc: 0.6553 -> 0.6012 | Hurt |
| vary final-state camera view (Event to View) | CNN- -G · TRANCE | Acc: 0.4419 -> 0.2300 | Hurt |
| increase transformation length from 1 to 4 steps | CNN- -G · TRANCE Event | Acc: 92%, 55%, 23%, and 8% | Hurt |
| add pretrained CLIP weights (frozen) (ablation) | ViT-B/32-T · TRANCO | EMR: 0.0378 -> 0.0881 | Improved |
| add pretrained CLIP weights (frozen) (ablation) | ViT-B/32-T · TRANCO tiny candidates | EMR@100: 0.1210 -> 0.1986 | Improved |
| further finetune pretrained CLIP weights (ablation) | ViT-B/32-T · TRANCO | EMR: 0.0881 -> 0.0748 | Hurt |
Closest papers by meaning
- CodeDance: A Dynamic Tool-integrated MLLM for Executable Visual Reasoning (CVPR 2026 · similarity 0.89)
- Multi-step Visual Reasoning with Visual Tokens Scaling and Verification (NeurIPS 2025 · similarity 0.89)
- STAR-R1: Multi-View Spatial TrAnsformation Reasoning by Reinforcing Multimodal LLMs (CVPR 2026 · similarity 0.88)
- CLiViS: Unleashing Cognitive Map through Linguistic-Visual Synergy for Embodied Visual Reasoning (CVPR 2026 · similarity 0.88)
- CoFFT: Chain of Foresight-Focus Thought for Visual Language Models (NeurIPS 2025 · similarity 0.88)
- UniVR: Thinking in Visual Space for Unified Visual Reasoning (NeurIPS 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.
