Aditya Rana

I am a computer vision and machine learning engineer at I recently completed my Master's in Computer Vision at Universitat Autònoma de Barcelona in a joint program with 3 more universities. During this time, I also worked at IRI, CSIC-UPC and Kognia Sports under the supervision of Francesc Moreno-Noguer and Antonio Rubio Romano which developed into my Master's thesis.

Prior to that, I finished my undergrad at BITS Pilani, India majoring in Electrical and Electronics Engineering, where I was also a teaching assistant for the course Neural Networks and Fuzzy Logic . For my bachelor's thesis, I interned at the Computer Vision Center's ADAS group where I was supervised by Dr. Joan Serrat and Dr. Antonio Lopez to work on neural image compression.

Mail  /  CV  /  Blog  /  GitHub  /  LinkedIn  /  Twitter

profile photo

I'm interested in algorithms for visual recognition (object recognition, localization and segmentation), generative models (especially for image compression) and machine learning with graphs.

graph_from_position Event Detection in Football using Graph Convolutional Networks
Aditya Rana, Dr. Antonio Rubio and Dr. Francesc Moreno
Master's Thesis, 2021
Thesis was developed under an NDA and the code was integrated into Kognia's platform

Football tracking data can be difficult to model because of lack of clarity on how to order players in a sequence and how to handle missing objects of interest. We show how to model the ball-player tracking data as graphs, and how to process them using graph convolutional networks. We focus on the task of event detection and present the results for different types of graph convolutional layers and losses that can be used to model the temporal context present around actions like goals, free-kicks, fouls etc.

importance_map_png Content Weighted Image Compression for Distributed Data Collection
Aditya Rana, Dr. Joan Serrat and Dr. Antonio Lopez
Bachelor's Thesis, 2020
code / bibtex

We show how to design lossy image compression models for optimizing data collection strategies from on road autonomous vehicles. The end goal is to gain advantage of the exposure of these vehicles to vast number of geographically diverse environments for training better detection and segmentation models.

Design courtesy of Jon Barron