The problem Virtual Trial Room - A site to try clothes online solves

Since the covid crisis , people have a major urge to go out for shopping but due to the covid-19 pandemic people are afraid to go outside and markets are empty. This is a huge loss for shopkeepers and the fashion industry. It is not that people dont want to buy clothes they are afraid to go out in markets, and online fashion stores don't have the same experience as physical stores. Primarily due to the fact that people can't try on clothes.
In this project we try to mitigate this problem. We build a website where customer can upload his/her photo and select a garment, the inbuilt Machine Learning model would then output customers's picture wearing the selected garment. We try to create a virtual trial room experience.
This technology can also be scaled as an API to be used with many popular online Fashion stores. This technology can help online fashion stores to boost sales. It can also help local shopkeepers, we can build an app in which the local shopkeeper can register his shop. The shopkeeper can then upload the garments to this app. Customers can browse through these garments and try them via our API. If the customer likes something then he can order it to delivered.
We can fill the gap between interested customers and Online/Physical clothing shops. This will stimulate demand in the Fashion Industry and help in the economic recovery. Its a WIn-WIn situation for customers as well sellers

Challenges we ran into

Deployment on sites, we are planning it to upload it on cloud but time constraints isn't allowing us . So we are using local host to demonstrate which is giving good results. We need to collect more data to give good results on the scale of it's potential we are on 8/10 but if we had more time , we can train our model more on data and would've given us photorealistic results.

Another Challenge we faced was that the clothes transfer Generative model was not robust enough. We used the https://github.com/JDAI-CV/Down-to-the-Last-Detail-Virtual-Try-on-with-Detail-Carving model since it is the latest and one of the best open source implementations of clothes transfer. But this Deep Learning model was mainly trained on pictures of women. Therefore the model performed poorly on images of men. We could have finetuned the pretrained model on pictures of men and more Indian people but it was not possible given the time constraints.