---
title: "BanglaWild: An In-the-Wild Bengali Scene Text Recognition Benchmark for OCR and Vision-Language Models"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2608.03884v1"
permalink: "https://feed.astralor.com/items/itm_e927b258a7751066547f2819"
published: "2026-08-04T16:20:53.000Z"
score: 77
channel: "ai"
tags: ["arXiv","BANGLAWILD","LoRA","Scene Text Recognition","Multilingual OCR"]
---

# BanglaWild: An In-the-Wild Bengali Scene Text Recognition Benchmark for OCR and Vision-Language Models

## AI commentary

> What stands out most is that this paper puts conventional OCR and VLMs on the same exam, then uses fine-grained error taxonomy to overturn a long-held assumption—conjuncts are not the main bottleneck in Bengali; visual mis-recognition is. If you work on low-resource scene text recognition or want to assess fine-grained VLM capabilities, this benchmark and error taxonomy are highly worth studying.

## At a glance

In-the-wild Bengali scene text recognition remains largely unmeasured: existing resources target handwritten documents or constrained sign-board parsing, report only aggregate edit-distance metrics, and evaluate either conventional OCR or VLMs, never both on the same data. This paper introduces BANGLAWILD, a benchmark of 2,535 Bengali scene text images with verbatim gold transcriptions, categorical axes, diagnostic attributes, and orthographic standard forms. Evaluating 15 VLMs and 3 conventional OCR systems under multiple prompting strategies, LoRA fine-tuning, and LLM-as-a-Judge evaluation, the study reveals a persistent gap where larger models do not outperform smaller ones in the same family. Visual mis-recognition accounts for ~60% of errors, while conjunct-related errors contribute under 2%, challenging a long-standing assumption in Bengali OCR research.

## Key points

- BANGLAWILD contains 2,535 Bengali scene text images with verbatim transcriptions, two categorical axes, four diagnostic attributes, and orthographic standard forms.
- Evaluates 15 VLMs and 3 conventional OCR systems with three prompting strategies and LoRA fine-tuning on 6 open-source models.
- Reveals that larger models in the same family do not outperform smaller ones; visual mis-recognition accounts for ~60% of errors, while conjunct-related errors are under 2%.
- Prompt language mainly affects cross-script drift; LoRA reduces catastrophic failures in weak models but does not raise the ceiling for competent ones.

## Full text

In-the-wild Bengali scene text recognition is largely unmeasured: existing resources target handwritten documents or constrained sign-board parsing, report only aggregate edit-distance metrics, and evaluate either conventional OCR or VLMs, never both on the same in-the-wild data. To address this gap, we introduce BANGLAWILD, a benchmark of 2,535 Bengali scene text images, each paired with a verbatim gold transcription, two categorical axes, four diagnostic attributes, and an orthographically standard form where the in-image text deviates from canonical spelling. We evaluate fifteen VLMs and three conventional OCR systems under three prompting strategies, fine-tune 6 open-source models with LoRA, and complement edit-distance metrics with an LLM-as-a-Judge evaluation. Our results reveal a persistent gap in which larger models within the same family do not outperform smaller ones. Our fifteen-class error taxonomy shows that visual mis-recognition accounts for ~60% of errors in the strongest systems, while conjunct-related errors contribute under 2%, challenging a long-standing assumption in Bengali OCR research; the same visual dominant profile also holds across architectures, including the one conventional baseline that reads Bengali reliably. Prompt language mainly affects cross-script drift and LoRA reduces catastrophic failures in weak models without lifting the ceiling on already competent ones. Code and data will be publicly released.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_e927b258a7751066547f2819