Uipath Deoldify

A UiPath RPA bot to automatically convert your old B&W videos to colorized format. This project is based on the Deoldify by Jason Antic and uses Google Colab for code execution.

Requirements

To run this project you need to have

  1. Uipath installed along with the required extension.
  2. Chrome Browser
  3. UiPath Web Automation Extension -> link
  4. Logged in into your Google account in chrome. (to use colab)

Instructions

  1. Open the project.json file in UiPath.
  2. Click the Run icon

Note: If converting a video using a url, you will have to enter a url from these supported sites only. The colorized video will be stored in your downloads folder with name video.mp4.

Inspiration

I heard about Deoldify during the Facebook F8 conference and tried colorizing some photos in Google Colab. The results I saw were pretty amazing. I instantly decided to try and turn this into a mobile app as it would help people not familiar with Colab, to try it out. However I soon realized that the model is too heavy to run on a phone. Also, running it on a cloud server would cost me a lot of dollars. An alternative was to somehow leverage Google Colab's resources itself in the background, but provide an easy interface on the front end for the user. Having played around with UiPath a bit the past year, I decided to build an RPA bot for it and also make it my entry to this hackathon.

What I learnt

Nothing is as easy as it looks. What I thought would be a piece of cake, turned out to be a hell hole. I had to spend the majority of my time correctly detecting when google Colab had completed the processing and safely extracting the output out of the platform. I had to find alternate routes all the time, because there would always be something that would break the next time I ran the bot.

How I built it

With lots of patience. I didn't want to use any extra libraries or tools for it. My goal was to make this project run on any system with google chrome installed. Saving the user from installing a ton of data science libraries and frameworks, was my main motive for this project. All the processing is done remotely on the cloud.

Challenges you faced

Majority of the work was done in decoding Colab CSS. I had to know exactly when the video was done processing and and how I could get its output url. I had to redo some flows multiple times because of some complex Colab CSS and iframes.

Built With

Share this project:

Updates