Swin-Tiny-Patch4-Window7-224-Finetuned-Skin-Cancer
| Property | Value |
|---|---|
| Base Model | microsoft/swin-tiny-patch4-window7-224 |
| License | Apache 2.0 |
| Accuracy | 72.75% |
| Training Framework | PyTorch 1.11.0 |
What is swin-tiny-patch4-window7-224-finetuned-skin-cancer?
This is a specialized image classification model fine-tuned on the HAM10000 skin cancer dataset using Microsoft's Swin Transformer architecture. The model can classify seven different types of skin conditions including Actinic keratoses, Basal cell carcinoma, Benign keratosis-like lesions, Dermatofibroma, Melanocytic nevi, Melanoma, and Vascular lesions.
Implementation Details
The model utilizes a Swin Transformer architecture with patch size 4 and window size 7, trained using the Adam optimizer with a learning rate of 5e-05. Training was conducted with a batch size of 128 (32 base batch size with 4 gradient accumulation steps) for one epoch, incorporating a linear learning rate scheduler with 10% warmup ratio.
- Training Loss: 0.6911
- Validation Loss: 0.7695
- Final Accuracy: 0.7275
- Implemented using Transformers 4.20.1 and PyTorch
Core Capabilities
- Multi-class skin condition classification
- Processing of medical imaging data
- Support for high-resolution dermatological images
- Inference on new skin lesion images
Frequently Asked Questions
Q: What makes this model unique?
This model combines the advanced Swin Transformer architecture with specialized medical imaging capabilities, specifically optimized for dermatological diagnosis. Its hierarchical feature processing makes it particularly effective for detecting subtle variations in skin lesions.
Q: What are the recommended use cases?
The model is designed to assist in preliminary screening of skin conditions and can be used as a supporting tool for dermatologists. However, it should not be used as the sole diagnostic tool, and all predictions should be verified by qualified medical professionals.
