swin-tiny-patch4-window7-224-finetuned-leukemia-08-2024

This model is a fine-tuned version of microsoft/swin-tiny-patch4-window7-224 on the imagefolder dataset. It achieves the following results on the evaluation set:

• Loss: 0.1282
• Accuracy: 0.9527

Model description

This model was developed to aid in the diagnosis of Leukemia. Leukemia is the cancer that most affects children between 4 and 10 years of age.

Intended uses & limitations

In this version, the images used for testing correspond to 10% of the images in the image dataset. Consider that the images in the dataset are originally:

• All: 7,272
• Hem: 3,389

Applying Data augmentation, we arrive at: 20,000 images for each segment. Totaling 36,000 images for training and 4,000 for testing.

Important Note:

The images used in the test may be similar to those used in training, which may cause overfit..

Training and evaluation data

DataSet: Dataset ISBI-2019 (ISIC 2019 Challenge)

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.328	0.9991	281	0.7951	0.6827
0.2832	1.9982	562	0.4021	0.8433
0.1886	2.9973	843	0.3305	0.8718
0.1789	4.0	1125	0.2242	0.9123
0.1269	4.9991	1406	0.1856	0.9315
0.0904	5.9982	1687	0.1282	0.9527
0.0754	6.9973	1968	0.1824	0.9377
0.0549	8.0	2250	0.2908	0.9105
0.0616	8.9991	2531	0.2961	0.9215
0.0502	9.9911	2810	0.2343	0.9345

Framework versions

• Transformers 4.44.2
• Pytorch 2.3.0+cpu
• Datasets 2.21.0
• Tokenizers 0.19.1

Author

Douglas Braga https://www.linkedin.com/in/douglas-braga-891a701/