Research map / Vision-language and multimodal / Multimodal LLMs
Molmo-Point: Better Pointing for VLMs with Grounding Tokens
Christopher Clark, Yue Yang, Jae Sung Park, Zixian Ma, Jieyu Zhang, Rohun Tripathi, Mohammadreza Salehi, Sangho Lee, Ranjay Krishna
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 grounding tokens to tracking model (ablation) | MolmoPoint-8B-Ablation · Molmo2Track overall | F1: 54.7 -> 59.3 | Improved |
| add grounding tokens to tracking model (ablation) | MolmoPoint-8B-Ablation · Molmo2Track overall | HOTA: 55.3 -> 59.3 | Improved |
| add MolmoPoint-Track data (ablation) | MolmoPoint-8B-Ablation · Molmo2Track Misc | F1: 30.3 -> 37.9 | Improved |
| add rotary embeddings (ablation) | Molmo2-P-Ablation · Molmo2-VC | Correct: 56.8 -> 58.0 | Improved |
| add rotary embeddings (ablation) | Molmo2-P-Ablation · PixMo-Points | F1: 84.5 -> 85.2 | Improved |
| add no-more-points class (ablation) | Molmo2-P-Ablation · Molmo2-VC | Overcount: 10.3 -> 3.6 | Improved |
| add no-more-points class (ablation) | Molmo2-P-Ablation · PointBench | Avg: 66.6 -> 67.8 | Improved |
| add sorted point generation order (ablation) | Molmo2-P-Ablation · Molmo2-VC | Correct: 40.0 -> 58.0 | Improved |
| add sorted point generation order (ablation) | Molmo2-P-Ablation · PointBench | Avg: 71.2 -> 67.8 | Hurt |
| use grounding tokens instead of text coordinates on natural images | MolmoPoint-8B · PixMo-Points | F1: 85.2 -> 89.2 | Improved |
| use grounding tokens vs text coordinates for GUI fine-tuning | MolmoPoint-GUI-8B · ScreenSpot-Pro | accuracy: 52.3 -> 60.2 | Improved |
| add 64-crop test-time scaling | MolmoPoint-GUI-8B · ScreenSpot-Pro | accuracy: 60.2 -> 61.1 | Improved |
Closest papers by meaning
- Molmo and PixMo: Open Weights and Open Data for State-of-the-Art Vision-Language Models (CVPR 2025 · similarity 0.88)
- Pointing at Parts: Training-Free Few-Shot Grounding in Multimodal LLMs (CVPR 2026 · similarity 0.87)
- MonoVLM: Monocular 3D Visual Grounding with Vision Language Models (CVPR 2026 · similarity 0.86)
- PixFoundation 2.0: Do Video Multi-Modal LLMs Use Motion in Visual Grounding? (NeurIPS 2026 · similarity 0.86)
- Enhancing Part-Level Point Grounding for Any Open-Source MLLMs (CVPR 2026 · similarity 0.86)
- Grounding Everything in Tokens for Multimodal Large Language Models (CVPR 2026 · similarity 0.85)
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.
