Research map / 3D vision / 3D generation
Kestrel: 3D Multimodal LLM for Part-Aware Grounded Description
Mahmoud Ahmed, Junjie Fei, Jian Ding, Eslam Mohamed Bakr, Mohamed Elhoseiny
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 query refinement levels (0 -> 2) in the segmentation decoder (ablation) | Kestrel · 3DCoMPaT-GrIn PaPGD | mIoU: 68.44 -> 86.70 | Improved |
| use Uni3D point encoder instead of PointBert (ablation) | Kestrel · 3DCoMPaT-GrIn PaPGD | mIoU: 77.05 -> 86.70 (with MLP projector) | Improved |
| use MLP projector instead of Q-Former (ablation) | Kestrel · 3DCoMPaT-GrIn PaPGD | mIoU: 74.50 -> 86.70 (with Uni3D) | Improved |
| add direct segmentation (DS) training data to grounded description data (ablation) | Kestrel · 3DCoMPaT-GrIn Direct Segmentation | mIoU: 49.97 -> 72.57 | Improved |
| add reasoning segmentation (RS) training data (ablation) | Kestrel · 3DCoMPaT-GrIn Reasoning Segmentation | mIoU: 52.03 -> 71.8 | Improved |
| increase direct segmentation data from 8K to 40K samples (ablation) | Kestrel · 3DCoMPaT-GrIn Grounded Description | mIoU: GD mIoU 86.07 -> 81.37; DS mIoU 78.78 -> 79.86 | Mixed |
| use Kestrel instead of SegPoint reimplementation | 3D MLLM segmentation grounding · 3DCoMPaT-GrIn Reasoning Segmentation | mIoU: 65.20 -> 71.80 | Improved |
| use 3D point input instead of 2D-view uplifting (Kestrel vs PARIS3D) | part grounding · 3DCoMPaT-GrIn Direct Segmentation | mIoU: 37.99 -> 78.78 | Improved |
| Kestrel zero-shot vs PartSTAD on PartNet-Mobility | Kestrel · PartNet-Mobility | overall accuracy: 65.0 vs 47.7 zero-shot, 63.9 few-shot | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- PARIS3D — PARIS3D: Reasoning-based 3D Part Segmentation Using Large Multimodal Model (ECCV 2024) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- 3D-PLOT-LLM: Part-Level Object Tokens for 3D Large Language Models (NeurIPS 2026 · similarity 0.88)
- Part-X-MLLM: Part-aware 3D Multimodal Large Language Model (ICLR 2026 · similarity 0.87)
- MMScan: A Multi-Modal 3D Scene Dataset with Hierarchical Grounded Language Annotations (NeurIPS 2024 · similarity 0.87)
- Pointing at Parts: Training-Free Few-Shot Grounding in Multimodal LLMs (CVPR 2026 · similarity 0.87)
- ReasonGrounder: LVLM-Guided Hierarchical Feature Splatting for Open-Vocabulary 3D Visual Grounding and Reasoning (CVPR 2025 · similarity 0.87)
- LIBA: Language Instructed Multi-granularity Bridge Assistant for 3D Visual Grounding (AAAI 2025 · similarity 0.87)
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.
