The international Silly-hacks competition is global hackathon where people from France, Indonetia etc participated. The task was to build something brilliant, yet silly aimed at just having fun. We built Silly-Cam winning us the competition. It uses machine learning, and generates viral meme videos with the face of the user instead.
Watch the video
Here is the submission that won us this Hackathon!
It's been recorded and dictated by my team mate, but the video sums up how the app worked :D
The idea
Deep fake is a really fun technology and has been around for a while. So I figured, why not use this and bring it to the phone of our users! The python code uses GPU, I added the code to pass in any image and it will morph it within the chosen Viral Video.
Deployment
Since the code was extremely complex and needed gpu to generate a video within seconds, I converted the code into a docker container and built a flask api so it can communicate outside python. Then I uploaded this container to Google cloud and and hosted my code there. We designed a flutter app and connected the app to my api, hosted in the cloud. Any user can enter a custom photo into the app and it'd send it to cloud, generate a video and send it back to the user!
Optimization and Innovation
Since generating a video almost took 15 minutes on CPU, I had to optmize the code a lot and performed multi-processing and other mechanims and refactored the code accordingly. Lastly I engaged GPU instead and brough run time down to about 25 seconds!
Technologies used and skills learned were:
- Tensorflow
- Python 3
- Flutter
- Optimization
- Google Cloud
- Deep-Fake
- Rest API
- Flask