isumenuka/medsiglip-nail-disease-classifier

MedSigLIP Nail Disease Classifier

Model Description

This model is fine-tuned from google/medsiglip-448 for nail disease classification. It can detect 7 different nail conditions with 51.10% accuracy.

Detected Conditions

1. Acral_Lentiginous_Melanoma
2. Healthy_Nail
3. Onychogryphosis
4. blue_finger
5. clubbing
6. pitting
7. psoriasis

Performance

• Accuracy: 51.10%
• Training Epochs: 8
• Image Size: 448x448

Usage

import torch
from transformers import AutoProcessor
from PIL import Image

# Load processor
processor = AutoProcessor.from_pretrained("YOUR_HF_USERNAME/medsiglip-nail-classifier")

# Load model
model = torch.load("pytorch_model.bin")
model.eval()

# Inference
image = Image.open("nail_image.jpg")
inputs = processor(images=image, return_tensors="pt")

with torch.no_grad():
    outputs = model(**inputs)
    predictions = outputs.logits.argmax(dim=-1)
    
print(f"Predicted class: {predictions.item()}")

Deployment to Vertex AI

1. Upload this model to HuggingFace Hub
2. In Google Cloud Vertex AI, navigate to Model Garden
3. Select "Import" → "From HuggingFace"
4. Enter your model repository URL
5. Deploy to get prediction endpoint

Training Details

• Base Model: google/medsiglip-448 (frozen)
• Classifier Architecture: 1152 → 768 → 512 → 256 → 7
• Optimizer: AdamW with OneCycleLR
• Data Augmentation: Extensive (rotation, flip, color jitter, etc.)

Limitations

• This model is for research purposes only
• Not approved for clinical diagnosis
• Should be used alongside professional medical evaluation

Citation

If you use this model, please cite:

@misc{medsiglip-nail-classifier,
  author = {Your Name},
  title = {MedSigLIP Nail Disease Classifier},
  year = {2026},
  publisher = {HuggingFace},
  howpublished = {\url{https://huggingface.co/YOUR_USERNAME/medsiglip-nail-classifier}}
}