Research map / Vision-language and multimodal / Multimodal LLMs
LIRA: Reasoning Reconstruction via Multimodal Large Language Models
Zhen Zhou, Tong Wang, Yunkai Ma, Xiao Tan, Fengshui Jing
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 confidence branch for candidate masks (ablation) | LIRA · ReasonRecon test | AP50: 29.30 -> 34.39 | Improved |
| use fine-tuned LLaVA-style reasoning segmentation module (final) instead of SEEM (ablation) | LIRA · ReasonRecon test | AP50: 11.00 -> 34.39 | Improved |
| use own 2D reasoning module instead of fine-tuned LISA (ablation) | LIRA · ReasonRecon test | AP50: 22.66 -> 34.39 | Improved |
| use own 2D reasoning module instead of fine-tuned LLaVA-Grounding (ablation) | LIRA · ReasonRecon test | AP50: 34.26 -> 34.39 | No clear effect |
| add text features to instance fusion (ablation) | LIRA · ReasonRecon test | AP50: 33.13 -> 34.39 | Improved |
| add bbox features to instance fusion (ablation) | LIRA · ReasonRecon test | AP50: 33.21 -> 34.39 | Improved |
| add voxel features to instance fusion (ablation) | LIRA · ReasonRecon test | AP50: 33.68 -> 34.39 | Improved |
| fuse multiple keyframes simultaneously within FBV (ablation) | LIRA · ReasonRecon test | AP50: 34.15 -> 34.39 | Improved |
| use ChatGPT-4o-mini instead of Llama3 in stage III (ablation) | LIRA · ReasonRecon test | AP50: 33.27 -> 34.39 | Improved |
| use ChatGPT-4o-mini instead of Qwen2.5 in stage III (ablation) | LIRA · ReasonRecon test | AP50: 33.58 -> 34.39 | Improved |
| use ChatGPT-4o instead of ChatGPT-4o-mini in stage III (ablation) | LIRA · ReasonRecon test | AP50: 34.39 -> 34.74 | No clear effect |
| use larger MLLM (LLaVA-13B) and ChatGPT-4o for reasoning | LIRA · ReasonRecon test | AP50: 34.39 -> 35.32 | Improved |
Closest papers by meaning
- VLM-3R: Vision-Language Models Augmented with Instruction-Aligned 3D Reconstruction (CVPR 2026 · similarity 0.89)
- Vid-LLM: A Compact Video-based 3D Multimodal LLM with Reconstruction–Reasoning Synergy (ICLR 2026 · similarity 0.87)
- MLLM-For3D: Adapting Multimodal Large Language Model for 3D Reasoning Segmentation (NeurIPS 2025 · similarity 0.86)
- Inst3D-LMM: Instance-Aware 3D Scene Understanding with Multi-modal Instruction Tuning (CVPR 2025 · similarity 0.86)
- Imagine3D-LLM: Teaching MLLMs to Imagine 3D Scenes Before Answering (NeurIPS 2026 · similarity 0.86)
- G$^2$VLM: Geometry Grounded Vision Language Model with Unified 3D Reconstruction and Spatial Reasoning (CVPR 2026 · similarity 0.86)
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.
