Research map / Vision-language and multimodal / Multimodal LLMs
Robin3D: Improving 3D Large Language Model via Robust Instruction Tuning
Weitai Kang, Haifeng Huang, Yuzhang Shang, Mubarak Shah, Yan Yan
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 adversarial instruction data (HROC) to benchmark training data (ablation) | Robin3D · ScanRefer | Acc@0.5: 45.3 -> 47.7 | Improved |
| add HOPE data (ablation) | Robin3D · Multi3DRefer | F1@0.5: 50.2 -> 52.5 | Improved |
| add PF-3DVG data (ablation) | Robin3D · ScanRefer | Acc@0.5: 45.3 -> 45.7 | No clear effect |
| add HROC data (ablation) | Robin3D · Scan2Cap | CIDEr@0.5: 73.6 -> 78.9 | Improved |
| add adversarial instruction data (ablation) | Robin3D · Scan2Cap | CIDEr@0.5: +8.9% | Improved |
| add adversarial instruction data (ablation) | Robin3D · ScanRefer / Multi3DRefer | accuracy / F1: +3.7% / +4.9% | Improved |
| add diverse instruction data (ChatGPT-rephrased styles) (ablation) | Robin3D · ScanRefer | accuracy: +5.3% | Improved |
| add Relation-Augmented Projector and sequence organization (post-vision order, ID wrapping) (ablation) | Chat-Scene baseline · ScanRefer / Multi3DRefer | grounding accuracy | Improved |
| train Robin3D on 1M RIG data instead of Chat-Scene benchmark-only data | Chat-Scene · Scan2Cap | CIDEr@0.5: 77.1 -> 87.2 | Improved |
| train Robin3D vs Chat-Scene on zero-target grounding | Chat-Scene · Multi3DRefer | F1@0.25: 57.1 -> 64.9 | Improved |
Closest papers by meaning
- WizardLM: Empowering Large Pre-Trained Language Models to Follow Complex Instructions (ICLR 2024 · similarity 0.87)
- 3D-PLOT-LLM: Part-Level Object Tokens for 3D Large Language Models (NeurIPS 2026 · similarity 0.87)
- Enhancing NLU in Large Language Models Using Adversarial Noisy Instruction Tuning (AAAI 2025 · similarity 0.87)
- 3D-GRAND: A Million-Scale Dataset for 3D-LLMs with Better Grounding and Less Hallucination (CVPR 2025 · similarity 0.86)
- Mixture-of-Experts Meets Instruction Tuning: A Winning Combination for Large Language Models (ICLR 2024 · similarity 0.86)
- VLM-3R: Vision-Language Models Augmented with Instruction-Aligned 3D Reconstruction (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.
