DeepSeek OCR 2

The LLM that powers DeepSeek's ability to recognize various files and images is open-sourced by DeepSeek. Important Notes: 1. The difference between this application and the first version of the DeepSeek OCR application is that the second version uses a recognition pattern similar to human attention (switching from the CLIP to the Qwen VL). - Advantages: When recognizing cluttered images, it will recognize text more like a "human brain" automatically focusing on the key parts of the text, and the layout will also be similar to human reading order. - Disadvantages: The side effect is that it is prone to hallucinations, and the output is not as stable as the first version, therefore it is released as a separate application. 2. Accuracy of ordinary image recognition: Built-in OCR model in the AI Pod ≈ DeepSeek OCR > DeepSeek OCR 2 3. Speed ​​of ordinary image recognition: Built-in OCR model in the AI Pod >> DeepSeek OCR ≈ DeepSeek OCR 2 4. For scenarios requiring high-quality layout, you can try using DeepSeek OCR 2 or DeepSeek OCR, as their layout capabilities are superior to the built-in OCR model in the AI Pod.