Pulkit Kumar

I am persuing my PhD at University of Maryland, College Park, advised by Abhinav Shrivastava. My current research interests are in video understanding, mainly action understanding. I am also interested in pushing the boundries of open world vision.

Before starting my graduate studies, I was a senior data scientist at ParallelDots, where I worked in computer vision and speech recognition with its applications in market research. I was also a research associate under Dr. Anubha Gupta at IIIT, Delhi where I explored the applications of computer vision in medical imaging diagnostics.

I did my bachelors at Netaji Subhas Institute of Technology, New Delhi in Information Technology, where I was advised by Dr. Anand Gupta in my Bachelor Thesis Project. During my junior and senior year, I split my time studying at NSIT and interning with the amazing team at ParallelDots.

Email / CV / Google Scholar / LinkedIn

Recent News

[Jul 24] One paper accepted at ECCV 2024.
[Apr 24] One paper accpted at CVPR 2024 workshop.
[Feb 24] One paper accpeted at CVPR 2024.
[May 23] Co-organising second iteration of OBJ-DISC challenge in VPLOW workshop at CVPR'23.
[May 23] Co-organising FMDC challenge in VPLOW workshop at CVPR'23.
[Sep 22] Co-organining second iteration of DNOW at WACV'23
[May 22] Co-organising OBJ-DISC challenge at VPLOW workshop held in conjuction with CVPR'22.
[Apr 22] Accepted to intern with the Visual Dynamics team at Google Research for Summer 2023.
[Sep 21] Co-organising Dealing with Novlety in Open Worlds (DNOW) workshop at WACV 2022.
[Aug 21] Joined the PhD. program at UMD!
[May 21] Got my Masters in Computer Scinece at UMD!

Research

	Trajectory-aligned Space-time Tokens for Few-shot Action Recognition Pulkit Kumar, Namitha Padmanabhan, Luke Luo, Sai Saketh Rambhatla, Abhinav Shrivastava European Conference on Computer Vision (ECCV), 2024 Harnessing Point Tracking and DINO, to create trajectory-aligned tokens (TATs) to capture motion and semantic information for few-shot action recognition.
	Explaining the Implicit Neural Canvas (XINC): Connecting Pixels to Neurons by Tracing their Contributions Namitha Padmanabhan^, Matthew Gwilliam^,Pulkit Kumar , Shishira R Maiya, Max Ehrlich, Abhinav Shrivastava Computer Vision and Pattern Recognition (CVPR), 2024 XINC dissects Implicit Neural Representation (INR) models to understand how neurons represent images and videos and to reveal the inner workings of INRs.
	Deep Multimodal Learning for the Diagnosis of Autism Spectrum Disorder Michelle Tang, Pulkit Kumar, Hao Chen, Abhinav Shrivastava Journal of Imaging, 2020 Incorporating two functional imaging modalities in an automated end-to-end autism diagnosis system for extracting comprehensive picture of the neural activity, and thus allowing more accurate diagnoses.
	Prototypical metric transfer learning for continuous speech keyword spotting with limited training data Harhita Seth, Pulkit Kumar* , Muktabh M. Srivastava International Conference on Soft Computing Models in Industrial and Environmental Applications (SOCO), 2019* A novel few-shot technique of combining prototypical network's loss with the metric loss and using transfer learning to form prototypes of domain specific keywords for their detection in continous speech.
	LeukoNet: DCT-based CNN architecture for the classification of normal versus Leukemic blasts in B-ALL Cancer Simmi Mourya, Sonaal Kant, Pulkit Kumar, Ritu Gupta , Anubha Gupta In submission* A deep learning framework for classifying immature leukemic blasts and normal cells by fusing Discrete Cosine Transform (DCT) domain features extracted via CNN with the Optical Density (OD) space features.
	U-Segnet: Fully convolutional neural network based automated brain tissue segmentation tool Pulkit Kumar , Pravin Nagar , Chetan Arora , Anubha Gupta International Conference on Image Processing (ICIP), 2018 A hybrid of SegNet and U-Net architecture for segmentation of Grey Matter, White Matter and Cerebrospinal Fluid in brain MRI.
	Boosted Cascaded Convnets for Multilabel Classification of Thoracic Diseases in Chest Radiographs Pulkit Kumar* , Monika Grewal, Muktabh M. Srivastava International Conference Image Analysis and Recognition (ICIAR)*, 2018 Combining boosting and cascading with DenseNets to detect all the pathologies in the Chest X-Ray 8 dataset.
	RADNET: Radiologist level accuracy using deep learning for hemorrhage detection in CT scans Monika Grewal, Muktabh M. Srivastava, Pulkit Kumar* , Srikrishna Varadarajan* International Symposium of Biomedical Imaging (ISBI), 2018 A Deep Learning model combining DenseNets with attention and LSTMs to detect haemorrhage from brain CT scans which matches the accuracy of senior radiologists.
	Anatomical labeling of brain CT scan anomalies using multi-context nearest neighbor relation networks Srikrishna Varadarajan, Muktabh M. Srivastava, Monika Grewal, Pulkit Kumar** Poster in International Symposium of Biomedical Imaging (ISBI), 2018 Used multi-context feature embeddings from a pre-trained VGG model with nearest neighbours to train RelationNets for anatomic labelling in brain CT Scans.
	A Big Data Analysis Framework Using Apache Spark and Deep Learning Anand Gupta, Hardeo Thakur, Ritvik Shrivastava, Pulkit Kumar, Sreyashi Nag International Conference of Data Mining (ICDM) workshop on Data Science and Big Data Analytics (DSBDA), 2017 A cascaded approach to predict the approval of H-1B visas on factors such as qualification, salary, location of job etc.

He is a great guy.