Yumna Javaid

Academic

Career

Verbatim text

Obstructy 
Obstructy is an android native mobile application which provides standard video editing features 
packaged with the salient speciality of “obstruction removal”. It is aimed at young adult users, users 
who spend significant time on social media, or users who are photography enthusiasts; but it can 
effectively be useful for anyone who uses a smart phone. The app is based on a recent 
breakthrough 
Deep 
Learning 
paper 
titled 
“Learning 
to 
see 
through 
obstructions” 
[https://www.youtube.com/watch?v=ICr6xi9wA94&t=18s]. Our app’s core function is obstruction 
removal from short videos; it takes in a short video clips and returns de-obstructed frames. This is 
done by employing 4 interconnected tensorflow deep learning models which remove unwanted 
obstructions (fences, raindrops, reflections only) using depth and angular information from short 
videos by identifying background and foreground layer, estimating their pixels’ flow over time, and 
reconstructing both layers/flows separately. A major goal thus is to deploy the model as a docker 
instance onto SageMaker, Amazon Web Services, or even deploy it on a simple AWS instance as 
proof of concept. Our app’s interface was inspired after humble UX analyses such as from 
competitor apps and also from informal user analyses. Our final UI leverages the Google Materials 
library.In addition to the deep learning model, our app also manipulates bit streams in mp4 videos 
for video editing, and provides the following features: 
- 
Obstruction removal (short videos to images/frames) 
- 
Video trimming 
- 
Video filters 
- 
Video audio manipulation; removing audio, adding music Video speed control 
 
 
 
Technology Used: 
Java, Python, Tensorflow, C/++ (FFMPEG),  
Android, Android Studio, Amazon Web Services 
Supervisor Name: 
Mr. Shoaib Mehboob 
Group Members:   
M. Huzaifa (i17 - 0305) 
Yumna Javaid (i17- 0215) 
                                                                                                                              Raja Salman Tariq (i17 - 0322)

AI enrichment