google-vit-large-finetuned

This Vision Transformer model is a fine-tuned version of Google's "vit-large-patch16-224" model.

This model has been fine-tuned with a custom dataset as a finishing project for an academic study.

The aim of the project is to develop a model that achieves high consistency with a limited amount of data. The study uses a dataset consisting of breast cancer images of varying resolutions.

The dataset contains 780 MRI images with a total of 3 classes (benign, malignant, normal), separated into train and test.

Distributions of images:

train:

• benign: 350
• malignant: 168
• normal: 106

test:

• benign: 87
• malignant: 42
• normal: 27

Since the size of the images varies, the images were scaled down to the resolution specified by Google for the model (224x224) and given to the model for fine-tuning.

Arguments used in fine-tuning:

trainArgs = TrainingArguments(
    save_strategy="epoch",
    evaluation_strategy="epoch",
    learning_rate=2e-5,
    per_device_train_batch_size=10,
    per_device_eval_batch_size=4,
    num_train_epochs=40,
    weight_decay=0.01,
    load_best_model_at_end=True,
    metric_for_best_model="accuracy",
    logging_dir='logs',
    remove_unused_columns=False,
)