Research map / Vision-language and multimodal / Multimodal LLMs
Groma: Localized Visual Tokenization for Grounding Multimodal Large Language Models
Chuofan Ma, Yi Jiang, Jiannan Wu, Zehuan Yuan, Xiaojuan Qi
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 |
|---|---|---|---|
| replace CLIP with DINOv2 as image encoder for localization (ablation) | DDETR head on frozen backbone · COCO detection | AP: 32.4 -> 38.9 (336x336) | Improved |
| increase input resolution from 336 to 448 for DINOv2 (ablation) | DDETR head on DINOv2 · COCO detection | AP: 38.9 -> 43.6 | Improved |
| unfreeze the LLM during instruction finetuning (ablation) | Groma · Visual Genome region captioning | CIDEr: 148.0 -> 158.4 | Improved |
| unfreeze the LLM during instruction finetuning (ablation) | Groma · REC benchmarks | average accuracy: 84.02 -> 86.52 | Improved |
| merge every four neighboring image tokens before the LLM (ablation) | Groma · REC benchmarks | average accuracy: 86.55 -> 86.47 | No clear effect |
| localized visual tokenization instead of LLM coordinate output or external modules | Groma · LVIS-Ground | AR: 16.8 -> 28.8 vs Ferret | Improved |
| use DINOv2 image tokens instead of CLIP-aligned tokens in the MLLM | Groma · LLaVA-Bench (COCO) | conversation and reasoning score: Conversation 82.6 vs LLaVA 85.4; Reasoning 88.8 vs 92.1 | Hurt |
| frozen-LLM Groma versus prior region models (ablation) | Groma (frozen LLM) · Visual Genome region captioning | CIDEr: 148.0 vs GPT4RoI 145.2 | Improved |
Later papers that compared with this method
6 later accepted papers report Groma-7B as a baseline in their results tables; 6 of them report a better number on most of the results they share with it.
- Rex-Thinker: Grounded Object Referring via Chain-of-Thought Reasoning (ICLR 2026 · better on 2 of 2 shared results)
- Falcon: Functional Assembly and Language for Compositional Reasoning in X-ray (ECCV 2026 · better on 1 of 1 shared result)
- ClawMachine: Learning to Fetch Visual Tokens for Referential Comprehension (ICLR 2025 · better on 6 of 6 shared results)
- LLaVA-ST: A Multimodal Large Language Model for Fine-Grained Spatial-Temporal Understanding (CVPR 2025 · better on 2 of 2 shared results)
- DynRefer: Delving into Region-level Multimodal Tasks via Dynamic Resolution (CVPR 2025 · better on 1 of 1 shared result)
- Text4Seg: Reimagining Image Segmentation as Text Generation (ICLR 2025 · better on 1 of 1 shared result)
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.
- Ferret — Ferret: Refer and Ground Anything Anywhere at Any Granularity (ICLR 2024) · better on 3 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
- Your Large Vision-Language Model Only Needs A Few Attention Heads For Visual Grounding (CVPR 2025 · similarity 0.90)
- Grounding Everything in Tokens for Multimodal Large Language Models (CVPR 2026 · similarity 0.90)
- Grounding Multimodal Large Language Models to the World (ICLR 2024 · similarity 0.89)
- GroundingME: Exposing the Visual Grounding Gap in MLLMs through Multi-Dimensional Evaluation (CVPR 2026 · similarity 0.89)
- GLaMM: Pixel Grounding Large Multimodal Model (CVPR 2024 · similarity 0.88)
- GeM-VG: Towards Generalized Multi-image Visual Grounding with Multimodal Large Language Models (AAAI 2026 · similarity 0.88)
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.
