The aim of the project is to provide hands-on experience on Computational Photography topics together with MLops paragidms. Throughout the project, students are intended to have freedom on topic/backend technology/mobile platform selection.
Topics of interest
Super Resolution
Image Dehazing
General Denoising
Motion Deblurring
Exposure Correction
Deep HDR
De-raining
Depth Estimation
De-focus
Image Coloring
More topics are welcomed but get ask the instructor first !
Roadmap
Do read papers and seek at least one github repositories involving source code of recent approaches (2019+)
Train the model or find a pre-trained model belonging to the subject of interest
Create a solution in which the client is an Android/iOS app which uploads a test image to a server for inferencing and retrieving the resultant image
The server can be AWS, Azure or even your own PC and it functions as the model inference server
The client should accept an image and send it to the model server in order to retrieve the result back!
Performing the inference natively on Android/Apple device (i.e. No server requirement) (+30 points). However, this requires a special treatment which could be quantization and you must show and describe the work of your own.
Inference stage should be completed under 1 minute
You can benefit from Flask or Streamlit as the ML backend server. However, you cannot use a browser based mobile app to demonstrate your project. The mobile app must be native.
A final report including all the key findings, architecture, run-time measurement and build procedure is a mandatory (25 points)
Any custom optimization for the extra speedup that you do it by yourself (e.g. quantization of the trained model) will be rewarded with 10 points
Any cheating, if detected, will be penalized with 0 points. However, you can, for sure, exhange ideas and ask help for building mobile app
Important Notes & Evaluation
The project can be done in pairs. Whole source code for the project must include different subfolders for different phases (i.e. deep model, backend service, mobile app).
The project must be presented by at least one student at the end of the semester. The week will be announced in the class.