mmabdpr/Deep-Learning-Course-Project-OCR

Usage

Evaluation

To run the saved model on custom image data you have to follow these steps:

• Train the model and save a checkpoint with the following format:
  • A dictionary that has at least one key named model which points to the state_dict of the pytorch model (CRNN48)
• Alternatively, you can download a trained model from this link.
• Place your images in a directory.
• In the same directory create a json file named labels.json with the following format:

   [
       {
           "file": "path/to/an/image.png",
           "label": "12359129"
       },
       {
           "file": "path/to/another/image.png",
           "label": "1235123159"
       }
   ]

• Edit the path variables in evaluate.py accordingly.
  • eval_labels should point to the labels.json.
  • checkpoint should point to the checkpoint file.
• Run src.evaluate module.
• Results are saved in a file named eval_results.json besides the labels.json.

Notes

• If any of the images are too narrow in width, an exception will occur.