vit-base-patch16-224-in21k-leukemia

This model is a fine-tuned version of google/vit-base-patch16-224-in21k on the Leukemia Dataset hosted on kaggle https://www.kaggle.com/datasets/andrewmvd/leukemia-classification. It achieves the following results on the evaluation set:

• Train Loss: 0.3256
• Train Accuracy: 0.8795
• Validation Loss: 0.6907
• Validation Accuracy: 0.6848
• Epoch: 13

Model description

Google Vision Transormer (ViT). fine-tuned on the white blood cancer - Leukemia - dataset

Intended uses & limitations

This model was fine-tuned as a part of my project LeukemiaAI, a fully integrated pipeline to detect Leukemia.

Github Repo: https://github.com/MohammedSaLah-Eldeen/LeukemiaAI

Training hyperparameters

• training_precision: mixed_float16
• optimizer: { 'inner_optimizer': { 'module': 'keras.optimizers.experimental', 'class_name': 'SGD', 'config': { 'name': 'SGD', 'weight_decay': None, 'clipnorm': None, 'global_clipnorm': 1, 'clipvalue': None, 'use_ema': False, 'ema_momentum': 0.99, 'ema_overwrite_frequency': None, 'jit_compile': True, 'is_legacy_optimizer': False, 'learning_rate': { 'module': 'keras.optimizers.schedules', 'class_name': 'CosineDecay', 'config': { 'initial_learning_rate': 0.001, 'decay_steps': 896, 'alpha': 0.0, 'name': None, 'warmup_target': None, 'warmup_steps': 0 }, 'registered_name': None }, 'momentum': 0.9, 'nesterov': False }, 'registered_name': None }, 'dynamic': True, 'initial_scale': 32768.0, 'dynamic_growth_steps': 2000

}

Training results

Train Loss	Train Accuracy	Validation Loss	Validation Accuracy	Epoch
0.5007	0.7629	0.7206	0.6643	0
0.3958	0.8418	0.7137	0.6686	1
0.3578	0.8632	0.6998	0.6789	2
0.3377	0.8713	0.6899	0.6843	3
0.3274	0.8778	0.6869	0.6832	4
0.3261	0.8792	0.6880	0.6859	5
0.3257	0.8797	0.6906	0.6848	6
0.3255	0.8796	0.6896	0.6859	7
0.3256	0.8794	0.6901	0.6848	8
0.3258	0.8795	0.6867	0.6864	9
0.3258	0.8793	0.6896	0.6859	10
0.3256	0.8796	0.6871	0.6864	11
0.3255	0.8795	0.6897	0.6853	12
0.3256	0.8795	0.6907	0.6848	13

Framework versions

• Transformers 4.35.0
• TensorFlow 2.13.0
• Datasets 2.1.0
• Tokenizers 0.14.1