trocr-base-printed_captcha_ocr

This model is a fine-tuned version of microsoft/trocr-base-printed on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.1380
• Cer: 0.0075

Model description

This model extracts text from image Captcha inputs.

For more information on how it was created, check out the following link: https://github.com/DunnBC22/Vision_Audio_and_Multimodal_Projects/blob/main/Optical%20Character%20Recognition%20(OCR)/Captcha/OCR_captcha.ipynb

Intended uses & limitations

This model is intended to demonstrate my ability to solve a complex problem using technology. You are welcome to test and experiment with this model, but it is at your own risk/peril.

Training and evaluation data

Dataset Source: https://www.kaggle.com/datasets/alizahidraja/captcha-data

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
10.4464	1.0	107	0.5615	0.0879
10.4464	2.0	214	0.2432	0.0262
10.4464	3.0	321	0.1380	0.0075

Framework versions

• Transformers 4.22.1
• Pytorch 1.12.1
• Datasets 2.4.0
• Tokenizers 0.12.1