The project “Face Skin Analyzer” is a Computer Vision model developed by me and hosted on Hugging Face’s model repository. The model is designed to analyze text descriptions of facial problems and provide a classification of the problem type.
The model takes in a text input describing a facial problem, such as “I have red bumps on my cheeks,” and outputs a label indicating the type of problem, such as “Acne” or “Rosacea.” The model is trained on a dataset of facial problem descriptions and their corresponding labels, and uses machine learning techniques to learn patterns and features that distinguish between different problem types.
The “Face Problems Analyzer” model can be useful for healthcare professionals, dermatologists, and individuals seeking to better understand their facial problems. It can help with triage and prioritization of patients, as well as provide more accurate and personalized treatment recommendations based on the specific problem type. The model can also be integrated into healthcare systems or telemedicine platforms to improve access to care and reduce the burden on healthcare providers.
You can find my app at below link
3 Likes
It seems like the model takes an input image instead of text? Also on which dataset did you train your model on?
@tyqiangz I have web scraped images using duckduckgo web search using fast.ai library. Yes the input to app is an image.
Your app is pretty cool. It managed to detect that I might have some eyebags HAHA. How did you preprocess your data and train the model?
Your project is interesting. I am curious also about how you train your model.
thank you,
Katherine Moss
I created V2 of Face Skin Analyzer using streamlit. It now has increased number of skin problems and can recommend the solution to those skin problems using Amazon products.
Link to V2: www.pratikskarnik.com
pratikskarnik,
I just used your application. It is really nice. Congratulations!
It looks like the focus of your project is on the side of skin aesthetics, which is a lot to conquer!
Have you thought about broadening the scope to medical concerns? If so, you could do training and testing with the DDI image dataset, which concerns itself with skin cancer diagnosis. The images in this set are divided into two: malignant and benign. This dataset distinguishes itself because the diagnoses are histologically confirmed, and the images are diverse with respect to skin tone.
Here is the link to the research paper, announcing the creation of the dataset…
Disparities in dermatology AI performance on a diverse, curated clinical image set | Science Advances
Here is an image of the data holding the file names and labels…
Regards,
Katherine Moss
https://www.linkedin.com/pub/katherine-moss/3/b49/228
