Research map / Vision-language and multimodal / Multimodal LLMs
IMoRe: Implicit Program-Guided Reasoning for Human Motion Q&A
Chen Li, Chinthani Sugandhika, Yeo Keat Ee, Eric Peh, Hao Zhang, Hong Yang, Deepu Rajan, Basura Fernando
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 program-guided feature selection over multi-level ViT features (ablation) | IMoRe · Babel-QA | Overall accuracy: 0.607 -> 0.640 | Improved |
| condition reasoning on program functions instead of MAC soft attention over question words (ablation) | IMoRe · Babel-QA | Overall accuracy: 0.582 -> 0.607 | Improved |
| use implicit MAC reasoning instead of explicit NSPose reasoning, same ViT encoder (ablation) | motion Q&A · Babel-QA | Overall accuracy: 0.440 -> 0.582 | Improved |
| replace NSPose motion encoder with ViT (ablation) | NSPose · Babel-QA | Overall accuracy: 0.578 -> 0.440 | Hurt |
| use predicted programs instead of ground-truth programs (ablation) | IMoRe · Babel-QA | Overall accuracy: 0.640 -> 0.615 (IMoRe II); NSPose 0.578 -> 0.472 | Hurt |
| use program rather than question for text-aware motion feature (ablation) | IMoRe I · Babel-QA | Overall accuracy: 0.609 -> 0.598 | Hurt |
| add question type information in fusion (ablation) | IMoRe I · Babel-QA | Overall accuracy: 0.603 -> 0.609 | Improved |
| sample 224 random frames with five test runs (IMoRe II) instead of processing the full sequence in segments (IMoRe I) | IMoRe · Babel-QA | Overall accuracy: 0.609 -> 0.640 | Improved |
| use IMoRe instead of fine-tuned video-language models | IMoRe I · Babel-QA | Overall accuracy: 0.425 (Qwen-2.5-VL-3B) -> 0.609 | Improved |
| use IMoRe on HuMMan-QA body-part queries versus MotionCLIP-MLP | IMoRe · HuMMan-QA | Query body part accuracy: MotionCLIP-MLP 0.692 vs IMoRe I 0.665 | Mixed |
Closest papers by meaning
- Motion Question Answering via Modular Motion Programs (ICML 2023 · similarity 0.88)
- Motion-Grounded Video Reasoning: Understanding and Perceiving Motion at Pixel Level (CVPR 2025 · similarity 0.85)
- SOK-Bench: A Situated Video Reasoning Benchmark with Aligned Open-World Knowledge (CVPR 2024 · similarity 0.84)
- HSI-GPT2: A Dual-Granularity Large Motion Reasoning Model with Diffusion Refinement for Human-Scene Interaction (CVPR 2026 · similarity 0.84)
- TOMATO: Assessing Visual Temporal Reasoning Capabilities in Multimodal Foundation Models (ICLR 2025 · similarity 0.84)
- AvatarGPT: All-in-One Framework for Motion Understanding Planning Generation and Beyond (CVPR 2024 · similarity 0.84)
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.
