Hi there! I'm

Abhishek Kumar

Are you looking for a creative and focused person that loves to take initiatives? Look no further

I am a Computer Vision Engineer with a strong background in Computer Science and Machine Learning. I am proficient in developing and implementing computer vision algorithms for a variety of applications, including image recognition, object detection, and image segmentation. I build most frequently with Python, C++, PyTorch.

Resume

I'm Open to work from September 2023

I have a Masters degree in Computer Science with 3+ year experience in software development. I’m interested in building scalable and reliable deep computer vision applications. Previously, I worked with Dr. David Doermann to built analytic components to detect manipulations and inconsistencies in online news media for SemaFor (DARPA).

I am currently working on a project that can detect small object in manipulated images using a diffusion based segmentation method.

Here are a few technologies I've been working with recently:
  • Python
  • C++
  • Pytorch
  • NumPy
  • OpenCV
  • Sci-Kit Learn
  • AWS
  • Docker
  • Git
  • Linux

Experience

Computer Vision Researcher - IAD
Sept 2022 - Aug 2023

Buffalo, New York

  • Built CNN model to detect tampered images by analyzing JPEG compression errors with ~ 94% accuracy.
  • Researched on constrained convolutions to localize manipulations in news images using noise and edge features.
  • Created object labelling module to label small objects within manipulated regions in images using YOLO.
  • Designed a baseline model to detect multimodal (image & text) inconsistency in online news articles.
  • Developed text transformer tool that performs controlled text replacements to create multimodal inconsistent data.
Teaching Assitant - CSE 701/702 - University At Buffalo
June 2022 - Aug 2022

Buffalo, New York

  • Guided students to develop deep learning projects for video analysis in sports domain for CSE 701/702.
  • Conducted classes, graded assignments, and reviewed students technical presentations.
Software Engineer - Infosys
Feb 2018 - Aug 2020

Bhubaneswar, India

  • Developed task delegation system for PLM application in Python to manage production workflows for CUMMINS.
  • Built REST API in Flask to parse delegation data from JSON files received in hourly batches.
  • Automated batch jobs with robotic process automation reducing transfer failure rate by 20%.
  • Designed Pricing System for Fleet Guard parts in Java to ensure accurate and reliable data management.
  • Implemented efficient shell scripts for data extraction, parsing and transfer across 6 pricing interfaces.
  • Managed deployment and maintenance pipelines using Jenkins and Git ensuring system availability and reliability.

Education

2021 - 2023
Masters in Computer Science (Machine Learning and Computer Vision)
University at Buffalo, New York
GPA: 3.72 out of 4.0
  • Worked as a Computer Vision Research Assistant in the IAD lab under Dr.David Doermann.
  • Built analytic components for SemaFor (DARPA) as part of the UB-SRI team.
  • Researched and wrote my thesis on : Forensic Systems to Detected Manipulated News Media.
  • Was a Teaching Assistant for graduate level courses (CSE 701/702), sports video analytics at University at Buffalo.
  • Guided students to develop deep learning projects for video analysis.
  • Courses : Deep Learing, Computer Vision and Image Processing, Machine Learning, Information Retrieval, Neurosymbolic AI (seminar), Distributed Systems, Analysis and Design of Algorithms.

Extracurricular Activities

  • Ranked 3rd in UB Hack (org. by MLH) for our Decentralized News Platform
  • Ranked 7th in Soccernet 2022 competition for our work on soccer player re-identification in broadcast videos.
2013 - 2017
Bachelor of Technology in Mechanical Engineering
SRM University, Chennai
GPA: 3.94 out of 4
  • Researched on carbon nano-tube infused phase change materials to create efficient cooling fluids.
  • Relevant Courses: Robotics, Probability and Statistics, Linear Algebra, Numerical Methods, Matlab, Advanced Calculus.

Extracurricular Activities

  • Won 2nd Prize in the annual tech fest (Aaruush'14) for our Infrared based line follower robot.
  • Active member of the photography club.

Projects

Temporal Action Spotting
Temporal Action Spotting
Implemented a transformer based model with multiscale flow and RGB features to classify actions and identify temporal boundaries for 17 action classes. Achieved 52 % mAP and ranked 5th in Soccernet competition.
Click here to know more!
Text-to-Image using Residual GANs
Text-to-Image using Residual GANs
Developed a Deep Residual GAN network to create images from latent noise. Used BIRD CUB-200 dataset to train the model.
Click here to know more!
Image Denoiser
Image Denoiser
Developed a CNN model in PyTorch that uses residual learning on deep CNNs to remove noise from iamges.
Click here to know more!
Player Re-Identification
Player Re-Identification
Designed a dual branch deep learning model to re-identify soccer players across multiple camera viewpoints depicting the same action during the game. Fused appearence and body part features from a subnet of OpenPose using Compact bilinear pooling to get the fused features.
Click here to know more!
Splice Detector
Splice Detector
A Deep Convolutional Neural Network model to detect manipulated images using JPEG compression analysis.
Click here to know more!

Get In Touch

Drop an email for any questions. My inbox is always open !
Email