AtlasOCR: Building the First Open-Source Darija OCR Model with Vision Language Models

TL;DR

AtlasOCR is the first open-source OCR model for Darija, leveraging a fine-tuned 3B Vision Language Model trained on synthetic and real data, achieving state-of-the-art results.

Contribution

This work introduces AtlasOCR, the first open-source Darija OCR model using a 3B parameter VLM with innovative dataset curation and efficient fine-tuning strategies.

Findings

Achieves state-of-the-art performance on Darija OCR benchmarks.

Demonstrates robustness and generalization across Darija and Arabic OCR tasks.

Outperforms larger models in accuracy and efficiency.

Abstract

Darija, the Moroccan Arabic dialect, is rich in visual content yet lacks specialized Optical Character Recognition (OCR) tools. This paper introduces AtlasOCR, the first open-source Darija OCR model built by fine-tuning a 3B parameter Vision Language Model (VLM). We detail our comprehensive approach, from curating a unique Darija-specific dataset leveraging both synthetic generation with our OCRSmith library and carefully sourced real-world data, to implementing efficient fine-tuning strategies. We utilize QLoRA and Unsloth for parameter-efficient training of Qwen2.5-VL 3B and present comprehensive ablation studies optimizing key hyperparameters. Our evaluation on the newly curated AtlasOCRBench and the established KITAB-Bench demonstrates state-of-the-art performance, challenging larger models and highlighting AtlasOCR's robustness and generalization capabilities for both Darija and standard Arabic OCR tasks.