Learning to Zoom Efficiently with a Contrastive Curriculum

• URL: http://arxiv.org/abs/2609.03206v1
• Date: Wed, 02 Sep 2026 22:43:54 GMT
• Title: Learning to Zoom Efficiently with a Contrastive Curriculum
• Abstract summary: Using a zoom-in tool is an important foundational part of modern visual agents.<n>We propose a new intrinsic reward for learning tool use in MLLMs without the need for additional labels or warm-start SFT.<n>Our approach is competitive while being more efficient.
• Score: 52.710726359379265
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Using a zoom-in tool is an important foundational part of modern visual agents, because it allows to efficiently handle tasks involving high-resolution images. Most previous methods need an extensive warm-start supervised fine-tuning phase for teaching models zoom-in. We show that this is not necessary by proposing a new intrinsic reward for learning tool use in MLLMs without the need for additional labels or warm-start SFT. Our InfoNCE-style reward uses a curriculum of increasingly hard negative tool calls as a contrastive training signal. Empirical experiments on $V^*$, HRBench and MME-RealWorld show that our approach is competitive while being more efficient. When used as a drop-in replacement for SFT, we even outperform all baselines. To directly measure the zoom-in ability of models, we further introduce the scalable synthetic Muffin&Chihuahua (M&C) dataset. Each image consists of a grid with every cell either showing a muffin or chihuahua. Leveraging the M&C dataset's unique region of interest labels, we find that recall is the metric that most strongly correlates the zoom-in region with final task performance. Our model and code for reproduction is publicly available under https://github.com/UKPLab/emnlp2026-zoom-in

Related papers

• GeoEyes: On-Demand Visual Focusing for Evidence-Grounded Understanding of Ultra-High-Resolution Remote Sensing Imagery [69.05066425853326]
  "thinking-with-images" paradigm enables multimodal large language models (MLLMs) to actively explore visual scenes via zoom-in tools.<n>This is essential for ultra-high-resolution (UHR) remote sensing VQA, where task-relevant cues are sparse and tiny.<n>We propose GeoEyes, a training framework consisting of (1) a cold-start SFT dataset, UHR Chain-of-Zoom (UHR-CoZ), which covers diverse zooming regimes, and (2) an agentic reinforcement learning method, AdaZoom-GRPO, that explicitly rewards evidence gain and answer improvement during zoom
  arXiv  Detail & Related papers  (2026-02-15T15:50:55Z)
• Zooming without Zooming: Region-to-Image Distillation for Fine-Grained Multimodal Perception [43.08943307183693]
  Region-to-Image Distillation transforms zooming from an inference-time tool into a training-time primitive.<n>We show that our models achieve leading performance across multiple fine-grained perception benchmarks.
  arXiv  Detail & Related papers  (2026-02-12T12:00:35Z)
• Zoom in, Click out: Unlocking and Evaluating the Potential of Zooming for GUI Grounding [71.97466930670936]
  Grounding is a fundamental capability for building graphical user interface (GUI) agents.<n>In this paper, we investigate zoom as a strong yet underexplored prior to GUI grounding, and propose a training-free method, ZoomClick.<n> Experiments demonstrate that our method significantly boosts the performance of both general vision-language and specialized GUI grounding models.
  arXiv  Detail & Related papers  (2025-12-05T18:39:12Z)
• Thinking with Programming Vision: Towards a Unified View for Thinking with Images [23.596757163808906]
  We show that even state-of-the-art MLLMs are surprisingly brittle, showing significant performance degradation on images with simple orientation changes or natural corruptions.<n>We propose CodeVision, a flexible and scalable code-as-tool framework where the model generates code as a universal interface to invoke any image operation.
  arXiv  Detail & Related papers  (2025-12-03T12:44:15Z)
• Adaptive Masking Enhances Visual Grounding [12.793586888511978]
  We propose IMAGE, Interpretative MAsking with Gaussian radiation modEling, to enhance vocabulary grounding in low-shot learning scenarios. We evaluate the efficacy of our approach on benchmark datasets, including COCO and ODinW, demonstrating its superior performance in zero-shot and few-shot tasks.
  arXiv  Detail & Related papers  (2024-10-04T05:48:02Z)
• Pink: Unveiling the Power of Referential Comprehension for Multi-modal LLMs [49.88461345825586]
  This paper proposes a new framework to enhance the fine-grained image understanding abilities of MLLMs. We present a new method for constructing the instruction tuning dataset at a low cost by leveraging annotations in existing datasets. We show that our model exhibits a 5.2% accuracy improvement over Qwen-VL and surpasses the accuracy of Kosmos-2 by 24.7%.
  arXiv  Detail & Related papers  (2023-10-01T05:53:15Z)
• MOCA: Self-supervised Representation Learning by Predicting Masked Online Codebook Assignments [72.6405488990753]
  Self-supervised learning can be used for mitigating the greedy needs of Vision Transformer networks. We propose a single-stage and standalone method, MOCA, which unifies both desired properties. We achieve new state-of-the-art results on low-shot settings and strong experimental results in various evaluation protocols.
  arXiv  Detail & Related papers  (2023-07-18T15:46:20Z)
• Noise Self-Regression: A New Learning Paradigm to Enhance Low-Light Images Without Task-Related Data [86.68013790656762]
  We propose Noise SElf-Regression (NoiSER) without access to any task-related data.<n>NoiSER is highly competitive in enhancement quality, yet with a much smaller model size, and much lower training and inference cost.
  arXiv  Detail & Related papers  (2022-11-09T06:18:18Z)
• MAGNeto: An Efficient Deep Learning Method for the Extractive Tags Summarization Problem [0.0]
  We study a new image annotation task named Extractive Tags Summarization (ETS) The goal is to extract important tags from the context lying in an image and its corresponding tags. Our proposed solution consists of different widely used blocks like convolutional and self-attention layers.
  arXiv  Detail & Related papers  (2020-11-09T11:34:21Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.