Research map / 3D vision / 3D generation
3D-VCD: Hallucination Mitigation in 3D-LLM Embodied Agents through Visual Contrastive Decoding
Makanjuola Adekunmi Ogunleye, Eman Abdelrahman, Ismini Lourentzou
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 3D-VCD contrastive decoding | 3D-LLM · 3D-POPE (Random) | Yes rate (%): 99.81 -> 75.15 | Improved |
| add 3D-VCD contrastive decoding | 3D-LLM · 3D-POPE (Random) | accuracy (%): 50.07 -> 67.99 | Improved |
| add 3D-VCD contrastive decoding | 3D-LLM · 3D-POPE (Popular) | accuracy (%): 49.94 -> 54.00 | Improved |
| add 3D-VCD contrastive decoding | 3D-LLM · 3D-POPE (Adversarial) | Yes rate (%): 99.94 -> 87.82 | Improved |
| add 3D-VCD contrastive decoding | Qwen-14B-Instruct · HEAL (Distractor Injection) | CHAIR C_S (%): 16.45 -> 5.00 | Improved |
| add 3D-VCD contrastive decoding | Qwen-14B-Instruct · HEAL (Distractor Injection) | CHAIR C_O (%): 4.13 -> 3.55 | Improved |
| add 3D-VCD contrastive decoding | Llama-3-8B-Instruct · HEAL (Distractor Injection) | CHAIR C_S (%): 9.49 -> 12.43 | Hurt |
| add 3D-VCD contrastive decoding | Qwen-14B-Instruct · HEAL (Scene-Task Contradiction) | CHAIR C_O (%): 53.9 -> 1.5 | Improved |
| use geometric noise 0.05 instead of 0.01 (ablation) | 3D-VCD · 3D-POPE (Random) | accuracy (%): 63.86 -> 67.65 | Improved |
| use geometric noise 0.05 instead of 0.45 (ablation) | 3D-VCD · 3D-POPE (Random) | accuracy (%): 63.79 -> 67.65 | Improved |
| increase contrastive strength alpha from 0.5 to 1.0 (ablation) | 3D-VCD Low-SemSub-Geom · 3D-POPE (Random) | accuracy (%): 64.39 -> 67.99 | Improved |
| use low instead of high semantic substitution (ablation) | 3D-VCD alpha=1.0 · 3D-POPE (Random) | accuracy (%): 64.58 -> 67.99 | Improved |
Closest papers by meaning
- Mitigating Object Hallucinations in Large Vision-Language Models through Visual Contrastive Decoding (CVPR 2024 · similarity 0.91)
- CODE: Contrasting Self-generated Description to Combat Hallucination in Large Multi-modal Models (NeurIPS 2024 · similarity 0.91)
- SECOND: Mitigating Perceptual Hallucination in Vision-Language Models via Selective and Contrastive Decoding (ICML 2025 · similarity 0.91)
- Learning to Decode Against Compositional Hallucination in Video Multimodal Large Language Models (ICML 2026 · similarity 0.91)
- Hallo3D: Multi-Modal Hallucination Detection and Mitigation for Consistent 3D Content Generation (NeurIPS 2024 · similarity 0.91)
- Reducing Hallucinations in Large Vision-Language Models via Latent Space Steering (ICLR 2025 · similarity 0.90)
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.
