Fashion Recommendr

Discover Your Wardrobe!!!

Inspiration:

I had to go through an awkward phase of fashion in my life where I did not know what clothes looked good on me.

What it does

Uses computer vision to find out the body measurements of a person from their webcam footage or pictures. Then it uses these measurements to show clothes recommended by the model.

How I built it

Using python backend and flutter frontend. In python, I initially used mediapipe, a python library, to identify body joint points. Later, I used deeplabv3, an image segmentation model, to find the outline of the person in the image. Using the points from mediapipe, the algorithm draws a line on the axis of the shoulder and hip points that extends across the image. I then find out the intersections with the contour drawn from the deeplabv3 model and the mediapipe lines. This gives the length in pixels, using a pixel to cm ratio, I find out the actual length in cm. After this, I use Google's AI model, gemini, and ask it for the clothing recommendations. Finally, I run these items through a web scrapping program and show the results on flutter.

Accomplishments that I'm proud of

Found an efficient method to find body measurements. In my case, I integrated Image segmentation and Posture detection. Furthermore, made an extremely aesthetically pleasing website. Lastly, used algorithmic thinking to build the recommendation model.

What I learned

I learned about Image segmentation and how it works, using prompts using API(gemini). Along with this, I got knowledge about extracting information from an e-commerce scrapping algorithm.

What's next for Fashion Recommendr

Reducing user friction
Use a Smaller language model for faster and more personalised results
Use a more efficient program than the current e-commerce scrapping program. The new program will search the whole internet instead of only amazon.