Test-Time Scaling for VLMs

VLM: Vision-Language Model

Test-time scaling for vision-language models studies how to spend additional computation at inference time to improve answers about images, video, and visual tasks. Common approaches generate and select multiple answers, search or verify reasoning steps, revisit visual evidence, or adapt image resolution and reasoning effort to task difficulty. These methods are evaluated on settings such as visual question answering, multimodal reasoning benchmarks, long-video understanding, and embodied planning. A central tension is that more computation does not always help: gains depend on model quality, visual grounding, decoding format, and how fairly methods are compared under matched compute budgets.

Latest papers

23 papers4 papers in the last four weeks, with none the four weeks before. 0.0% of all new papers.

MAGIC: Margin-gated Contrastive Debiasing with Evidence Restoration for VLM Test-time Robustness Scaling
Oct 23, 2026Shunshun Gu, Enze Zhang, Ming Pei +2VLM Robustness Test-Time Scaling for VLMs
University of Wisconsin–Madison · University of North Carolina at Chapel Hill
ReTAC: Residual Test-Time Agentic Computation for Video Spatial Reasoning
Oct 12, 2026Yike Wu, Keyan Chen, Qixiong Wang +6Test-Time Scaling for VLMs Video Reasoning
Nanyang Technological University · Xiaohongshu Inc. · Zhejiang University of Technology +1
Architectural Sampling: Test-Time Scaling via Computational Diversity in Frozen Vision-Language Models
Oct 1, 2026Akshit Singh, Shyam Marjit, Wei Lin +2Test-Time Scaling for VLMs Large Vision-Language Models
Dartmouth College · IISc Bangalore · Xero
Token-Disentangled Latent Test-Time Scaling for Vision-Language Reasoning
Sep 28, 2026Hao-Xuan Ma, Yihao Liu, Yutao Sun +8Test-Time Scaling for VLMs Test-Time Scaling
Qwen Business Unit of Alibaba · School of Artificial Intelligence, Nanjing University · National Key Laboratory for Novel Software Technology, Nanjing University +5
It's the Decoding Format, Not the Perturbation: Auditing Consistency-Based Selection for Vision-Language Test-Time Scaling
Aug 2, 2026Puzhuo Zheng, Hasan KurbanVLM Evaluation Self-Consistency Decoding
College of Science and Engineering, Hamad Bin Khalifa University, Doha, Qatar
No Training, Better Flights: Test-Time Scaled VLMs for UAV Navigation
Jul 21, 2026Feinan Cheng, Dongliang Xu, Wenli Nong +4Vision-Language Models UAV Navigation
Shandong University Weihai, China · China University of Petroleum (East China) Qingdao, China · MBZUAI Abu Dhabi, United Arab Emirates
Test-Time Scaling for Small VLMs on Multilingual Visual MCQ
Jul 10, 2026Spiros Baxevanakis, Peng-Jian YangVisual Question Answering Test-Time Scaling for VLMs
University of Amsterdam, Science Park 904, 1098 XH Amsterdam, The Netherlands
On Test-Time Scaling for Vision-Language Models
Jun 27, 2026Fawaz Sammani, Tzoulio Chamiti, Nikos DeligiannisVLM Evaluation Test-Time Scaling for VLMs
ETRO Department, Vrije Universiteit Brussel, Belgium · imec, Kapeldreef 75, B-3001 Leuven, Belgium
DIRECT: When and Where Should You Allocate Test-Time Compute in Embodied Planners?
Jun 10, 2026Jadelynn Dao, Milan Ganai, Yasmina Abukhadra +7Test-Time Scaling for VLMs Inference-Time Scaling
Stanford University · University of Waterloo · NVIDIA
AVIS: Adaptive Test-Time Scaling for Vision-Language Models
Jun 10, 2026Ahmadreza Jeddi, Minh Ngoc Le, Amirhossein Kazerouni +8Visual Token Pruning Test-Time Adaptation of VLMs
AI Center-Toronto, Samsung Electronics · University of Toronto · Vector Institute +1
Diversity Matters: Revisiting Test-Time Compute in Vision-Language Models
May 29, 2026Yijie Tong, Yifan Hou, Shaobo Cui +2VLM Evaluation Vision-Language Models
ETH Zürich · Shanghai Jiao Tong University · EPFL
Test-Time Hinting for Black-Box Vision-Language Models
May 13, 2026Kaihua Hou, Abhijith Varma Mudunuri, Jiaxing Qiu +3Visual Question Answering VLM Adaptation
UC Berkeley · UCSF · University of Virginia +1
Test-time Scaling over Perception: Resolving the Grounding Paradox in Thinking with Images
Apr 13, 2026Zheng Jiang, Yiming Chen, Nan He +4Visually Grounded Reasoning Test-Time Scaling for VLMs
Tsinghua University, Beijing, China · Beijing University of Technology, Beijing, China
Visual Enhanced Depth Scaling for Multimodal Latent Reasoning
Apr 12, 2026Yudong Han, Yong Wang, Zaiquan Yang +3Latent Visual Reasoning Continuous Latent Reasoning
Beijing Institute of Technology · AMAP, Alibaba Group · City University of Hong Kong +2
VideoAtlas: Navigating Long-Form Video in Logarithmic Compute
Mar 18, 2026Mohamed Eltahir, Ali Habibullah, Yazan Alshoibi +3Long-Video Understanding Video Representation Learning
King Abdullah University of Science and Technology (KAUST), Thuwal, Saudi Arabia · Department of Computer Science, King Khalid University (KKU), Abha, Saudi Arabia · Department of Computer Science, Edge Hill University, Ormskirk, England
Linking Perception, Confidence and Accuracy in MLLMs
Mar 12, 2026Yuetian Du, Yucheng Wang, Rongyu Zhang +5Multimodal Large Language Models VLM Calibration
Zhejiang University · University of Michigan · Alibaba Group +1
GeoSolver: Scaling Test-Time Reasoning in Remote Sensing with Fine-Grained Process Supervision
Mar 10, 2026Lang Sun, Ronghao Fu, Zhuoran Duan +3VLM Reasoning Process Reward Models
College of Computer Science and Technology, Jilin University, Changchun 130012, China · Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education Jilin University
Scaling Test-Time Robustness of Vision-Language Models via Self-Critical Inference Framework
Mar 8, 2026Kaihua Tang, Jiaxin Qi, Jinli Ou +2VLM Robustness Test-Time Scaling for VLMs
Tongji University, China · Computer Network Information Center, CAS, China · University of Chinese Academy of Sciences, China +1
VisRef: Visual Refocusing while Thinking Improves Test-Time Scaling in Multi-Modal Large Reasoning Models
Feb 27, 2026Soumya Suvra Ghosal, Youngeun Kim, Zhuowei Li +6Test-Time Scaling for VLMs Multimodal Reasoning
University of Maryland, College Park · Amazon · Physion Labs
Saliency-Aware Multi-Route Thinking: Revisiting Vision-Language Reasoning
Feb 18, 2026Mingjia Shi, Yinhan He, Yaochen Zhu +1VLM Reasoning Reasoning for VLMs
University of Virginia.
SPARC: Separating Perception And Reasoning Circuits for Test-time Scaling of VLMs
Feb 6, 2026Niccolo Avogaro, Nayanika Debnath, Li Mi +6Visual Question Answering Efficient VLM Inference
ETH Zürich · IBM Research · MIT-IBM Watson AI Lab.
CASHEW: Stabilizing Multimodal Reasoning via Iterative Trajectory Aggregation
Jan 12, 2026Chaoyu Li, Fei Tao, Pooyan FazliTest-Time Scaling for VLMs Test-Time Scaling
Arizona State University · NewsBreak
Understanding the Effects of Distractors on Reasoning Vision-Language Models
Nov 26, 2025Jiyun Bae, Hyunjong Ok, Sangwoo Mo +1VLM Reasoning Visual Question Answering
Pohang University of Science and Technology (POSTECH)

Oct 23, 2026J/K move · Enter open · S save

MAGIC: Margin-gated Contrastive Debiasing with Evidence Restoration for VLM Test-time Robustness Scaling

Shunshun Gu, Enze Zhang, Ming Pei, Hang Zhou+1

University of Wisconsin–Madison · University of North Carolina at Chapel Hill

VLM Robustness Test-Time Scaling for VLMs Vision-Language Models Contrastive Decoding Visual Evidence Sensitivity in VLMs

Open paper AskPDF arXiv

This paper studies the problem of test-time robustness scaling for large vision-language models (LVLMs), which aims to mitigate language bias during inference. Existing approaches typically adopt test-time scaling techniques such as contrastive decoding to improve inference performance. Despite the progress, the performance of these approaches is far from satisfactory due to intentional evidence removal during perturbations and redundant inference cost. Towards this end, we propose a novel approach named Margin-gated Contrastive Debiasing with Evidence Restoration (MAGIC). The core of our MAGIC is to evaluate the decision margins of LVLMs to selectively activate textual and visual perturbation branches for efficient debiased inference. In particular, our MAGIC first measures the prediction margins of crucial tokens, and applies a gating function to identify low-confidence responses that are susceptible to language biases. For selected samples, we integrate the token logits from multiple textual and image perturbation branches with the original ones for debiased decision. More importantly, we selectively restore the most informative corrupted views with sharp prediction distributions for ambiguous samples after debiasing, enabling evidence recovery for discriminative decision-making. We further utilize a calibration dataset to actively identify and prune ineffective branches during contrastive debiasing for cost reduction. Extensive experiments on benchmark datasets validate the effectiveness of the proposed MAGIC in comparison with extensive baselines.