Unlimited OCR DeepEncoder

Unlimited OCR DeepEncoder is a browser-ready vision encoder model exported in ONNX format, designed for tasks involving optical character recognition (OCR) and document parsing. The model represents the complete vision encoder component from the baidu/Unlimited-OCR stack, specifically from the DeepSeek-OCR family, and incorporates a DeepEncoder architecture that fuses SAM ViT-B and CLIP-L models with a linear projector. It is numerically verified to be identical to its PyTorch reference, ensuring consistency across platforms.

The tool is intended for scenarios where OCR and document parsing are required directly within the browser, eliminating the need for server-side processing. It accepts images as input and produces detected, boxed markdown output, enabling a streamlined, client-side OCR pipeline. The model is paired with a language model in GGUF format to run the full Unlimited-OCR pipeline entirely in a browser tab. html.

Unlimited OCR DeepEncoder is distributed under the MIT license, allowing for broad use and integration into web-based applications. The delivery is focused on browser environments, leveraging onnxruntime-web and webgpu for in-browser inference. This makes the model particularly suitable for developers and researchers looking to integrate OCR capabilities into web applications without relying on external servers or backend infrastructure.