| | --- |
| | license: apache-2.0 |
| | base_model: Qwen/Qwen2-VL-2B-Instruct |
| | tags: |
| | - dermatology |
| | - medical |
| | - vision-language-model |
| | - lora |
| | - indian-health |
| | - peft |
| | datasets: |
| | - marmal88/skin_cancer |
| | - pvlinhk/ISIC2019-full |
| | language: |
| | - en |
| | --- |
| | |
| | # DermaAI LoRA Adapter — Indian Skin Type Tuned |
| |
|
| | Fine-tuned LoRA adapter on top of `Qwen2-VL-2B-Instruct` for clinical dermatological diagnosis |
| | with a specific focus on **South Asian skin types (Fitzpatrick IV–VI)** and Indian treatment protocols. |
| |
|
| | ## Training Data |
| | - **HAM10000** (marmal88/skin_cancer): 10,015 dermoscopic images, 7 diagnostic categories |
| | - **ISIC 2019** (pvlinhk/ISIC2019-full): 25,000+ images, 8 categories |
| | |
| | ## Capabilities |
| | - Skin lesion morphological description |
| | - Differential diagnosis generation |
| | - Indian pharmacological management (IADVL guidelines) |
| | - Optimized for: Tinea, Psoriasis, Leprosy (Hansen's), Vitiligo, Melanoma, BCC |
| | |
| | ## Usage |
| | ```python |
| | from transformers import AutoProcessor, Qwen2VLForConditionalGeneration |
| | from peft import PeftModel |
| | |
| | model = Qwen2VLForConditionalGeneration.from_pretrained("Qwen/Qwen2-VL-2B-Instruct") |
| | model = PeftModel.from_pretrained(model, "hssling/derm-analyzer-adapter") |
| | processor = AutoProcessor.from_pretrained("hssling/derm-analyzer-adapter") |
| | ``` |
| | |
| | ## Disclaimer |
| | Research use only. Not a substitute for clinical evaluation by a qualified dermatologist. |
| | |
