Integrate AtlasOCR Model into KITAB-Bench OCR Evaluation Task

[Elma-dev]

• AtlasOCR Class Implementation: A new class AtlasOCR has been added under ocr-eval/models/ to encapsulate the model loading, input preparation, and prediction logic for the AtlasOCR model.
• eval.py Integration (ocr-eval/eval.py):
  • Import AtlasOCR: The AtlasOCR model is now imported from the models module.
  • get_model Function Update: The get_model function now includes a condition to return an instance of AtlasOCR when model_name is "atlasocr".
  • AVAILABLE_MODELS Update: (Implicitly, assuming AVAILABLE_MODELS is updated in init_.py to include "atlasocr".) This allows the atlasocr model to be selected via command-line arguments for evaluation.

This integration allows us to leverage the AtlasOCR model for improved OCR accuracy and performance within the KITAB-Bench framework and enables its evaluation alongside other OCR models.