About

I am a PhD candidate in Computer Science at the Stony Brook University, working with Prof. Michael S. Ryoo and Prof. Dimitris Samaras. My broad research goal is to create efficient computer vision systems, enabling accessibility and fairness, while regulating the negative impact on the world. I am particularly interested in topics in video domain, and work on improving parameter/compute efficiency, inference speed and label efficiency. Currently, I am exploring video representations with minimal supervision, and excited about attention mechanisms.

Prior to this, I worked as a Research Assistant at University of Moratuwa, Sri Lanka, under the supervision of Dr. Ranga Rodrigo. I completed my B.Sc. in 2018 specializing in Electronic & Telecommunication Engineering at University of Moratuwa, advised by Prof. Dileeka Dias and Dr. Prathapasinghe Dharmawansa.

Recent News

[Nov 2021] Our recent paper "SWAT: Spatial Structure Within and Among Tokens" is now on arxiv.
[Aug 2021] I had a great internship experience at Wormpex AI Research. Our paper "Self-supervised Pretraining with Classification Labels for Temporal Activity Detection" is now on arxiv.
[June 2021] One paper was presented at CVPR 2021.
[May 2021] This Summer, I will be interning at Wormpex AI Research, working on video representation learning with minimal supervision.
[Mar 2021] Our paper "Coarse-Fine Networks for Temporal Activity Detection in Videos" was accepted at CVPR 2021.
[Jan 2021] Two papers were presented at WACV 2021 and ICPR 2020.
[Aug 2020] Our paper "Exploiting the Redundancy in Convolutional Filters for Parameter Reduction" was accepted at WACV 2021.
[Jun 2020] Our paper "Feature-dependent Cross-Connections in Multi-Path Neural Networks" was accepted at ICPR 2020.
[May 2020] Our paper "On the Exact Outage Probability of 2x2 MIMO-MRC in Correlated Rician Fading" was presented at IEEE Wireless Communications and Networking Conference (WCNC) 2020 held virtually.

Preprints

SWAT: Spatial Structure Within and Among Tokens
Kumara Kahatapitiya, Michael S. Ryoo
Self-supervised Pretraining with Classification Labels for Temporal Activity Detection
Kumara Kahatapitiya, Zhou Ren, Haoxiang Li, Zhenyu Wu, Michael S. Ryoo
MS-TCT: Multi-Scale Temporal ConvTransformer for Action Detection
Rui Dai, Srijan Das, Kumara Kahatapitiya, Michael S. Ryoo, Francois Bremond
(Soon on arxiv)
StARformer: Transformer with State-Action-Reward Representations for Visual Reinforcement Learning
Jinghuan Shang, Kumara Kahatapitiya, Xiang Li, Michael S. Ryoo
(Soon on arxiv)
Swift: Adaptive Video Streaming with Layered Neural Codecs
Mallesham Dasari, Kumara Kahatapitiya, Samir Das, Aruna Balasubramanian, Dimitris Samaras
(Soon on arxiv)

Publications

Coarse-Fine Networks for Temporal Activity Detection in Videos
Kumara Kahatapitiya, Michael S. Ryoo
CVPR 2021
Exploiting the Redundancy in Convolutional Filters for Parameter Reduction
Kumara Kahatapitiya, Ranga Rodrigo
WACV 2021
Feature-dependent Cross-Connections in Multi-Path Neural Networks
Dumindu Tissera, Kasun Vithanage, Rukshan Wijesinghe, Kumara Kahatapitiya, Subha Fernando, Ranga Rodrigo
ICPR 2020
On the Exact Outage Probability of 2x2 MIMO-MRC in Correlated Rician Fading
Prathapasinghe Dharmawansa, Kumara Kahatapitiya, Saman Atapattu, Chintha Tellambura
WCNC 2020
Context-Aware Automatic Occlusion Removal
Kumara Kahatapitiya, Dumindu Tissera, Ranga Rodrigo
ICIP 2019
Low-power Step Counting Paired with Electromagnetic Energy Harvesting for Wearables
Kumara Kahatapitiya, Chamod Weerasinghe, Jinal Jayawardhana, Hiranya Kuruppu, Kanchana Thilakarathna, Dileeka Dias
ISWC 2018

Open-source Projects

X3D-Multigrid [source]
A PyTorch implementation for "X3D: Expanding Architectures for Efficient Video Recognition models" [CVPR2020] with "A Multigrid Method for Efficiently Training Video Models" [CVPR2020]. In contrast to the original repository by FAIR, this repository provides a simpler, less modular and more familiar structure of implementation for faster and easier adoptation.
Optimal Transport in NumPy [source]
This repository contrains a few Optimal Transport Algorithms implemented using NumPy, including "A Direct O(1/epsilon) Iteration Parallel Algorithm for Optimal Transport" [NeurIPS2019], "Computational Optimal Transport: Complexity by Accelerated Gradient Descent is better than by Sinkhorn's Algorithm" [PMLR2018] and "Lightspeed Computation of Optimal Transport" [NeurIPS2013].

Teaching

CSE327: Computer Vision - TA (Spring 2020)
CSE215: Foundations of Computer Science - TA (Fall 2019)

Website template from Justin Johnson. Thanks!