Train-time zooming distillation enables fine-grained multimodal perception without inference latency.
Multimodal Large Language Models (MLLMs) excel at broad visual understanding but still struggle with fine-grained perception, where decisive evidence is small and easily overwhelmed by global context. Recent "Thinking-with-Images" methods alleviate this by iteratively zooming in and out regions of interest during inference, but incur high latency due to repeated tool calls and visual re-encoding. To address this, we propose Region-to-Image Distillation, which transforms zooming from an inference-time tool into a training-time primitive, thereby internalizing the benefits of agentic zooming into a single forward pass of an MLLM. In particular, we first zoom in to micro-cropped regions to let strong teacher models generate high-quality VQA data, and then distill this region-grounded supervision back to the full image. After training on such data, the smaller student model improves "single-glance" fine-grained perception without tool use. To rigorously evaluate this capability, we further present ZoomBench, a hybrid-annotated benchmark of 845 VQA data spanning six fine-grained perceptual dimensions, together with a dual-view protocol that quantifies the global--regional "zooming gap". Experiments show that our models achieve leading performance across multiple fine-grained perception benchmarks, and also improve general multimodal cognition on benchmarks such as visual reasoning and GUI agents. We further discuss when "Thinking-with-Images" is necessary versus when its gains can be distilled into a single forward pass. Our code is available at https://github.com/inclusionAI/Zooming-without-Zooming.
Region-to-Image Distillation moves zooming from inference to training, enabling single-pass fine-grained perception without latency overhead.
Teacher models generate high-quality VQA data on micro-crops where fine details are unambiguous, then distill back to full images.
This approach achieves accuracy benefits of iterative zooming while maintaining real-time inference speed through intelligent data synthesis.
Referential ambiguity arises when region-specific questions are applied to full images, requiring careful distillation strategies.
The approach depends on high-quality teacher models and consensus filtering, which may limit scalability and data diversity.
Jingxuan Wei, Siyuan Li +22
The construction of World Models capable of learning, simulating, and reasoning about objective physical laws constitutes a foundational challenge in the pursuit of Artificial General Intelligence. Re...
Haifeng Wang, Hua Wu +436
In this report, we introduce ERNIE 5.0, a natively autoregressive foundation model desinged for unified multimodal understanding and generation across text, image, video, and audio. All modalities are...
Ailin Huang, Chengyuan Yao +91
We present STEP3-VL-10B, a lightweight open-source foundation model designed to redefine the trade-off between compact efficiency and frontier-level multimodal intelligence. STEP3-VL-10B is realized t...
Yu Wang, Yi Wang +5
As hubs of human activity, urban surfaces consist of a wealth of semantic entities. Segmenting these various entities from satellite imagery is crucial for a range of downstream applications. Current ...
Hongrui Jia, Chaoya Jiang +2
As Large Multimodal Models (LMMs) scale up and reinforcement learning (RL) methods mature, LMMs have made notable progress in complex reasoning and decision making. Yet training still relies on static...
