Shlok Mishra

I am a Ph.D. student at University of Maryland in the department of Computer Science. I am advised by Prof. David Jacobs working on problems in Computer Vision and Machine Learning. I also obtained an MS in Computer Science from University of Maryland, College Park.

I primarily work in Self-Supervised Learning and Generative models in Computer Vision. I have done multiple internships at Google. In Summer of 2023, I worked with Prof. Ira Kemelmacher-Shlizerman on Try-On diffusion. In Summer of 2022, I worked with Dilip Krishnan on combining Self-Supervised learning and Generative modelling. In Summer 2021 I worked with Christian Haene and Hossam Isack on Generative models. In Summer 2020, I worked with Kuntal Sengupta, Vincent Chu and Sofien Bouaziz on Spoof Detection. I have also done 4 student research internships and have been working in Google for last 2.5 years.

Email  /  CV  /  Google Scholar  /  Twitter  /  Github  /  Linkedin

I'm on the Job market for Industry positions.

profile photo
News

Research

I am interested in computer vision, machine learning. Specifically, my current research interests are in Self-Supervised learning, Generative models and building models which can do learn both generative and discriminative features.

PoseAction Hyperbolic Contrastive Learning for Visual Representations beyond Objects
Shlok Mishra* Songwei Ge* , Simon Kornblith, Chun-Liang Li, David Jacobs
CVPR 2023

We propose a Hyperbolic loss to better encode scene and objects.

arXiv code
PoseAction A simple, efficient and scalable contrastive masked autoencoder for learning visual representations
Shlok Mishra*, Joshua Robinson* , Huiwen Chang , David Jacobs , Aaron Sarna, Aaron Maschinot, Dilip Krishnan
preprint

We propose a simple and scalable contrastive masked autoencoder method.

arXiv code
HaLP HaLP: Hallucinating Latent Positives for Skeleton-based Self-Supervised Learning of Actions

Anshul Shah, Aniket Roy*, Ketul Shah*, Shlok Mishra, David Jacobs, Anoop Cherian, Rama Chellappa
CVPR 2023

We hallucinate latent postives for learning skeleton encoders without labels

paper
PoseAction Robust Contrastive Learning Using Negative Samples with Diminished Semantics
Songwei Ge*, Shlok Mishra, Chun-Liang Li , Chun-Liang Li, David Jacobs
Neurips 2021

We show texture based hard negative samples can imporve generalization of contrastive learning method.

paper code
ObjAwareCrop Object-Aware Cropping for Self-Supervised Learning
Shlok Mishra, Anshul Shah , Ankan Bansal, Abhyuday Jagannatha, Abhishek Sharma, David Jacobs, Dilip Krishnan
TMLR 12/2022

A novel cropping strategy for SSL on uncurated datasets

arXiv code
PoseAction Generating Annotated Datasets via Keypoints Conditioned StyleGAN

Shlok Mishra, Hossam Isack , Sergio Orts-Escolano, Luca Prasso, Rohit Pandey, Franziska Mueller, Abhimitra Meka, Jonathan Taylor, Dilip Krishnan , David Jacobs , Christian Haene.
Under Submission

We use StyleGAN to generate labelled data for keypoint estimation task.

AnisotropicImageNet Learning Visual Representations for Transfer Learning by Suppressing Texture
Shlok Mishra, Anshul Shah , Ankan Bansal, Abhyuday Jagannatha, Jonghyun Choi, Abhinav Shrivastava, Abhishek Sharma, David Jacobs
BMVC 2022

Anisotropic diffusion based augmentation to reduce texture bias in supervised and self-supervised approaches

paper
PoseAction Improved Presentation Attack Detection Using Image Decomposition
Shlok Mishra, Kuntal Sengupta, Wen-Sheng Chu , Max Horowitz-Gelb, Sofien Bouaziz , David Jacobs IJCB 2022 (ORAL)

We show albedo is a strong cue for spoof detection.

arXiv
PoseAction Pose and Joint-Aware Action Recognition
Anshul Shah , Shlok Mishra, Ankan Bansal, Jun-Cheng Chen, Rama Chellappa, Abhinav Shrivastava WACV 2022

We present a new model and loss for Pose-based action recognition

arXiv / video / code
Service

Serving as reviewer for major machine learning and computer vision conferences (CVPR, Neurips, ICLR, AAAI, ECCV, ICCV, etc) and journals (TMLR, TIP , CVIU, IJCV). In the past I have also served in NLP conferences (ACL, EMNLP).



Template Credits : Jon Barron