Course Project - ENGI 9805 Computer Vision

A web app for generating photo-realistic Super-Resolution Images/Videos based on SRGAN.

Try the demo.

Due to the hardware limitation of the server, the online version of the appliction can only process images with size smaller than 240 x 240.

If you want to test on larger images, please download the app and run it on your local machine.

Installation

Clone the repo

git clone git@github.com:ENGI9805-COMPUTER-VISION/Term-Project.git

cd into the project root folder

cd Term-Project

Create a conda virtual environment

conda create -n srgan_env python=3.6.8

and activate environment

conda activate srgan_env

Then you need to install the project dependencies

pip install -r requirements.txt

Usage

Run the app

streamlit run app.py

Available models

• netG epoch 100 upscale factor 2
• netG epoch 100 upscale factor 4
• netG epoch 100 upscale factor 8

Work Log

• Build the program
  • Machine learning resource management
  • Data ingestion and collection (Jason)
  • Setup & configure cloud computing environment for GPU training (Jason)
  • Model training (Jason, Luo)
  • Integrating trained model with web app（Ian)
  • Process Single Video (Ian)
  • Add statistic terms definition
  • Fix png file alpha channel error
  • Fix Show Benchmark Datasets error
  • Refactor code and add comments
• Deploy working demo on Heroku platform
  • Add startup file (Jason)
  • Inject necessary dependencies (Jason)
  • Fix deployment error (Jason)
  • Add Image size check
• Write project report
  • Choose a suitable template and fine-tune the layout & style (Jason)
  • Abstract (Jason)
  • Introduction (Jason)
  • Problem definition (Luo)
  • Proposed solution (Ian)
  • Results and discussion (Jason & Luo)
  • Conclusion (Ian)
  • Reference (Jason)
  • Proof read (Jason)
• Write presentation slides
  • Choose a suitable template and fine-tune the layout & style (Jason)
  • Background & Motivation (Jason)
  • Our Solution (Ian)
  • Result (Jason)
  • Demo (Jason)
  • Conclusion & Futrue Work (Jason)
  • References (Jason)
• Record presentation video
  • Background & Motivation (Jason)
  • Our Solution (Ian)
  • Result (Jason)
  • Demo (Ian)
  • Conclusion & Futrue Work (Jason)
  • References (Jason)
  • Video Editing (Luo)