Where Vision Becomes Text: Locating the OCR Routing Bottleneck in Vision-Language Models

• URL: http://arxiv.org/abs/2602.22918v1
• Date: Thu, 26 Feb 2026 12:06:02 GMT
• Title: Where Vision Becomes Text: Locating the OCR Routing Bottleneck in Vision-Language Models
• Authors: Jonathan Steinberg, Oren Gal,
• Abstract summary: Vision-language models (VLMs) can read text from images, but where does this optical character recognition (OCR) information enter the language processing stream?<n>We investigate the OCR routing mechanism across three architecture families using causal interventions.
• Score: 2.1942030377331245
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Vision-language models (VLMs) can read text from images, but where does this optical character recognition (OCR) information enter the language processing stream? We investigate the OCR routing mechanism across three architecture families (Qwen3-VL, Phi-4, InternVL3.5) using causal interventions. By computing activation differences between original images and text-inpainted versions, we identify architecture-specific OCR bottlenecks whose dominant location depends on the vision-language integration strategy: DeepStack models (Qwen) show peak sensitivity at mid-depth (about 50%) for scene text, while single-stage projection models (Phi-4, InternVL) peak at early layers (6-25%), though the exact layer of maximum effect varies across datasets. The OCR signal is remarkably low-dimensional: PC1 captures 72.9% of variance. Crucially, principal component analysis (PCA) directions learned on one dataset transfer to others, demonstrating shared text-processing pathways. Surprisingly, in models with modular OCR circuits (notably Qwen3-VL-4B), OCR removal can improve counting performance (up to +6.9 percentage points), suggesting OCR interferes with other visual processing in sufficiently modular architectures.

Related papers

• SimpleOCR: Rendering Visualized Questions to Teach MLLMs to Read [43.28273039987167]
  We introduce the Visualized-Question (VQ) setting, where text queries are rendered directly onto images.<n>Despite possessing strong OCR capabilities, models suffer a performance degradation of up to 12.7% in the VQ setting.<n>We propose SimpleOCR, a plug-and-play training strategy that imposes a structural constraint on the learning process.
  arXiv  Detail & Related papers  (2026-02-25T21:36:30Z)
• PsOCR: Benchmarking Large Multimodal Models for Optical Character Recognition in Low-resource Pashto Language [2.1540520105079697]
  We develop a synthetic Pashto OCR dataset, PsOCR, consisting of one million images annotated with bounding boxes at word, line, and document levels.<n>PsOCR covers variations across 1,000 unique font families, colors, image sizes, and layouts.<n>A benchmark subset of 10K images was selected to evaluate the performance of several LMMs, including seven open-source models.<n> Gemini achieves the best performance among all models, whereas among open-source models, Qwen-7B stands out.
  arXiv  Detail & Related papers  (2025-05-15T07:58:38Z)
• VISTA-OCR: Towards generative and interactive end to end OCR models [3.7548609506798494]
  VISTA-OCR is a lightweight architecture that unifies text detection and recognition within a single generative model.<n>Built on an encoder-decoder architecture, VISTA-OCR is progressively trained, starting with the visual feature extraction phase.<n>To enhance the model's capabilities, we built a new dataset composed of real-world examples enriched with bounding box annotations and synthetic samples.
  arXiv  Detail & Related papers  (2025-04-04T17:39:53Z)
• Benchmarking Vision-Language Models on Optical Character Recognition in Dynamic Video Environments [3.5936169218390703]
  This paper introduces an open-source benchmark for evaluating Vision-Language Models (VLMs) on Optical Character Recognition (OCR) tasks in dynamic video environments.<n>We present a curated dataset containing 1,477 manually annotated frames spanning diverse domains, including code editors, news broadcasts, YouTube videos, and advertisements.
  arXiv  Detail & Related papers  (2025-02-10T13:20:19Z)
• LOCR: Location-Guided Transformer for Optical Character Recognition [55.195165959662795]
  We propose LOCR, a model that integrates location guiding into the transformer architecture during autoregression. We train the model on a dataset comprising over 77M text-location pairs from 125K academic document pages, including bounding boxes for words, tables and mathematical symbols. It outperforms all existing methods in our test set constructed from arXiv, as measured by edit distance, BLEU, METEOR and F-measure.
  arXiv  Detail & Related papers  (2024-03-04T15:34:12Z)
• Vision-by-Language for Training-Free Compositional Image Retrieval [78.60509831598745]
  Compositional Image Retrieval (CIR) aims to retrieve the relevant target image in a database. Recent research sidesteps this need by using large-scale vision-language models (VLMs) We propose to tackle CIR in a training-free manner via Vision-by-Language (CIReVL)
  arXiv  Detail & Related papers  (2023-10-13T17:59:38Z)
• CoVR-2: Automatic Data Construction for Composed Video Retrieval [59.854331104466254]
  Composed Image Retrieval (CoIR) has recently gained popularity as a task that considers both text and image queries together. We propose a scalable automatic dataset creation methodology that generates triplets given video-caption pairs. We also expand the scope of the task to include composed video retrieval (CoVR)
  arXiv  Detail & Related papers  (2023-08-28T17:55:33Z)
• T-MARS: Improving Visual Representations by Circumventing Text Feature Learning [99.3682210827572]
  We propose a new data filtering approach motivated by our observation that nearly 40% of LAION's images contain text that overlaps significantly with the caption. Our simple and scalable approach, T-MARS, filters out only those pairs where the text dominates the remaining visual features.
  arXiv  Detail & Related papers  (2023-07-06T16:59:52Z)
• PP-OCRv3: More Attempts for the Improvement of Ultra Lightweight OCR System [11.622321298214043]
  PP-OCRv3 upgrades the text detection model and text recognition model in 9 aspects based on PP-OCRv2. Experiments on real data show that the hmean of PP-OCRv3 is 5% higher than PP-OCRv2 under comparable inference speed.
  arXiv  Detail & Related papers  (2022-06-07T04:33:50Z)
• Structured Multimodal Attentions for TextVQA [57.71060302874151]
  We propose an end-to-end structured multimodal attention (SMA) neural network to mainly solve the first two issues above. SMA first uses a structural graph representation to encode the object-object, object-text and text-text relationships appearing in the image, and then designs a multimodal graph attention network to reason over it. Our proposed model outperforms the SoTA models on TextVQA dataset and two tasks of ST-VQA dataset among all models except pre-training based TAP.
  arXiv  Detail & Related papers  (2020-06-01T07:07:36Z)

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.