vit-base-oxford-brain-tumor

This model is a fine-tuned version of google/vit-base-patch16-224 on the Mahadih534/brain-tumor-dataset dataset. It achieves the following results on the evaluation set:

• Loss: 0.5719
• Accuracy: 0.6923

Model description

This model is a fine-tuned version of google/vit-base-patch16-224, which is a Vision Transformer (ViT)

ViT model is originaly a transformer encoder model pre-trained and fine-tuned on ImageNet 2012. It was introduced in the paper "An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale" by Dosovitskiy et al. The model processes images as sequences of 16x16 patches, adding a [CLS] token for classification tasks, and uses absolute position embeddings. Pre-training enables the model to learn rich image representations, which can be leveraged for downstream tasks by adding a linear classifier on top of the [CLS] token. The weights were converted from the timm repository by Ross Wightman.

Intended uses & limitations

This must be used for classification of x-ray images of the brain to diagnose of brain tumor.

Training and evaluation data

The model was fine-tuned in the dataset Mahadih534/brain-tumor-dataset that contains 253 brain images. This dataset was originally created by Yousef Ghanem.

The original dataset was splitted into training and evaluation subsets, 80% for training and 20% for evaluation. For robust framework evaluation, the evaluation subset is further split into two equal parts for validation and testing. This results in three distinct datasets: training, validation, and testing

Training procedure/hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0003
• train_batch_size: 20
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 7

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
No log	1.0	11	0.5904	0.64
No log	2.0	22	0.5276	0.68
No log	3.0	33	0.4864	0.8
No log	4.0	44	0.4566	0.8
No log	5.0	55	0.4390	0.88
No log	6.0	66	0.4294	0.96
No log	7.0	77	0.4259	0.96

Framework versions

• Transformers 4.41.2
• Pytorch 2.3.0+cu121
• Datasets 2.19.2
• Tokenizers 0.19.1