Research map / Vision-language and multimodal / CLIP and zero-shot
Test-Time Computing for Referring Multimodal Large Language Models
Mingrui Wu, Hao Chen, Jiayi Ji, Xiaoshuai Sun, Zhiyuan Liu, Liujuan Cao, Ming-Ming Cheng, Rongrong Ji
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 Optim++ Adam optimizer in place of SGD (ablation) | ControlMLLM++ · ROC (box, test) | accuracy: 60.59 -> 66.93 | Improved |
| add answer-start token attention A(st) on top of Adam (ablation) | ControlMLLM++ · ROC (box, test) | accuracy: 66.93 -> 65.70 | Hurt |
| add layer selection on top of Adam (ablation) | ControlMLLM++ · ROC (box, test) | accuracy: 66.93 -> 65.70 | Hurt |
| add both A(st) and layer selection on top of Adam (ablation) | ControlMLLM++ · ROC (box, test) | accuracy: 66.93 -> 68.02 | Improved |
| add PromptDebias contrastive decoding (ablation) | ControlMLLM++ · ROC (box, test) | accuracy: 68.02 -> 71.19 | Improved |
| add early stopping to ControlMLLM (T=0 vs T=5) (ablation) | ControlMLLM · ROC (val) | accuracy: 57.00 -> 62.50 | Improved |
| lower EMA decay beta to 0.3 at alpha=400 (ablation) | ControlMLLM · ROC (val) | accuracy: 53.5 -> 62.5 | Improved |
| add ControlMLLM test-time optimization to LLaVA-1.5 | LLaVA-1.5 · ROC (box, test) | accuracy: 54.72 -> 60.59 | Improved |
| use ControlMLLM++ instead of ControlMLLM on Qwen2.5-VL-7B | Qwen2.5-VL-7B · ROC (box, test) | accuracy: 79.20 -> 79.20 | No clear effect |
| add ControlMLLM++ to Qwen2.5-VL | Qwen2.5-VL · RTC (box, test) | accuracy: 81.91 -> 88.23 | Improved |
| scale LLaVA-1.5 from 7B to 13B (vanilla) | LLaVA-1.5 · ROC (box, test) | accuracy: 54.72 -> 55.69 | No clear effect |
| add ControlMLLM++ to LLaVA-HR | LLaVA-HR · Screenshot referring description | CIDEr-D: 10.88 -> 66.96 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- Ferret-7B — Ferret: Refer and Ground Anything Anywhere at Any Granularity (ICLR 2024) · better on 2 of 5 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
- ControlMLLM: Training-Free Visual Prompt Learning for Multimodal Large Language Models (NeurIPS 2024 · similarity 0.95)
- Accelerating Multimodal Large Language Models via Dynamic Visual-Token Exit and the Empirical Findings (NeurIPS 2025 · similarity 0.92)
- POINTS-Long: Adaptive Dual-Mode Visual Reasoning in MLLMs (CVPR 2026 · similarity 0.91)
- Exploring the Transferability of Visual Prompting for Multimodal Large Language Models (CVPR 2024 · similarity 0.91)
- Look-Back: Implicit Visual Re-focusing in MLLM Reasoning (AAAI 2026 · similarity 0.90)
- Boosting Multimodal Large Language Models with Visual Tokens Withdrawal for Rapid Inference (AAAI 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.
