Research map / Vision-language and multimodal / Multimodal LLMs
IVQA-LD: Inclusive Multimodal Understanding for Population with Limb Deficiency
Yan Ke, Xin Shen, Jiaying Ying, Xin Li, Xin Yu
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces IVQA-LD, a limb-deficiency-aware vision-language dataset and benchmark for inclusive multimodal reasoning for people with limb differences.
Problem
People with limb differences face barriers to inclusive AI services, largely because structured, high-quality multimodal resources centered on disability contexts are lacking.
Approach
A limb-deficiency-aware body-centric learning and evaluation paradigm combining a large-scale limb-aware vision-language dataset with an evaluation benchmark for multimodal reasoning; the second component is cut off in the available abstract.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add stage-1 residual-limb grounding warm-up (S1) (ablation) | Qwen2.5-VL-7B SFT · IVQA-LD caption generation | CIDEr: 65.37 -> 76.15 | Improved |
| add stage-1 residual-limb grounding warm-up (S1) (ablation) | Qwen2.5-VL-7B SFT · IVQA-LD caption generation | METEOR: 26.04 -> 27.29 | Improved |
| add stage-1 residual-limb grounding warm-up (S1) (ablation) | Qwen2.5-VL-7B SFT · IVQA-LD guide generation | guide score: 4.54 -> 4.64 | Improved |
| add stage-2 grounding supervision (S2) on top of S1 (ablation) | Qwen2.5-VL-7B · IVQA-LD caption generation | CIDEr: 76.15 -> 79.20 | Improved |
| add stage-2 grounding supervision (S2) on top of S1 (ablation) | Qwen2.5-VL-7B · IVQA-LD caption generation | METEOR: 27.29 -> 27.33 | No clear effect |
| add stage-2 grounding supervision (S2) on top of S1 (ablation) | Qwen2.5-VL-7B · IVQA-LD guide generation | guide score: 4.64 -> 4.64 | No clear effect |
| add BSI over standard SFT | Qwen2.5-VL-7B · IVQA-LD visual grounding | Acc@0.5: 18.91 -> 26.34 | Improved |
| add BSI over standard SFT | Qwen2.5-VL-7B · IVQA-LD deficiency type | accuracy: 37.60 -> 52.60 | Improved |
| add BSI over standard SFT | Qwen2.5-VL-7B · IVQA-LD scene reasoning | accuracy: 83.00 -> 80.13 | Hurt |
| add BSI over standard SFT | Qwen2.5-VL-7B · IVQA-LD prosthesis interaction | accuracy: 67.33 -> 88.29 | Improved |
| add BSI over standard SFT | Qwen2.5-VL-7B · IVQA-LD WPA classification | accuracy: 63.80 -> 72.40 | Improved |
| fine-tune Qwen2.5-VL-7B on IVQA-LD vs zero-shot | Qwen2.5-VL-7B · IVQA-LD ICF classification | accuracy: 0.30 -> 73.11 | Improved |
Closest papers by meaning
- HandVQA: Diagnosing and Improving Fine-Grained Spatial Reasoning about Hands in Vision-Language Models (CVPR 2026 · similarity 0.86)
- Benchmarking Egocentric Multimodal Goal Inference for Assistive Wearable Agents (NeurIPS 2025 · similarity 0.84)
- Mordal: Automated Pretrained Model Selection for Vision Language Models (ICLR 2026 · similarity 0.83)
- OctoMed: Data Recipes for State-of-the-Art Multimodal Medical Reasoning (CVPR 2026 · similarity 0.83)
- Flex-MoE: Modeling Arbitrary Modality Combination via the Flexible Mixture-of-Experts (NeurIPS 2024 · similarity 0.83)
- Universal Skeleton Understanding via Differentiable Rendering and MLLMs (ICML 2026 · similarity 0.83)
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.
