Subhabrata Choudhury

Subhabrata Choudhury

PhD, Engineering Science

Visual Geometry Group

University of Oxford

United Kingdom

google scholar twitter twitter mail

University of Oxford

University of Oxford

PhD, Engineering Science

Advisor: Andrea Vedaldi

United Kingdom, 2019-2024

Max-Planck Institute for Informatics

Max-Planck Institute for Informatics

MSc Thesis

Advisor: Bernt Schiele

Germany, 2018-2019

Saarland University

Saarland University

MSc, Computer Science

Germany, 2017-2019

IIT Kharagpur

IIT Kharagpur

BTech (Hons), Metallurgical and Materials Engineering

India, 2009-2013

Meta Reality Labs Research

Meta Reality Labs Research Surreal

Research Scientist Intern

Host: Chris Sweeney

USA, 2022


The Curious Layperson: Fine-Grained Image Recognition without Expert Labels

IJCV 2023 | BMVC 2021 (Oral) [ Best Student Paper Award ]
pdfproject pagevideocodearXivBibTeX
Proposes a new problem of fine-grained image classification without expert annotations, by utilizing class agnostic non-expert descriptions and off-the-shelf expert corpus

Unsupervised Multi-Object Segmentation by Predicting Probable Motion Patterns

NeurIPS 2022
pdfproject pagecodearXivBibTeX
We learn to segment independent objects in still images by predicting regions that contain motion patterns likely to arise from such objects.

Guess What Moves: Unsupervised Video and Image Segmentation by Anticipating Motion

BMVC 2022 (Spotlight)
pdfproject pagevideocodearXivBibTeX
We use motion anticipation as a learning signal to train an image segmentation network to predict regions that likely contain similar optical flow patterns

Unsupervised Part Discovery from Contrastive Reconstruction

NeurIPS 2021
pdfproject pagevideocodearXivBibTeX
Unsupervised discovery of semantically meaningful parts of objects using appearance consistency cues and contrastive learning

Semantic Projection Network for Zero- and Few-Label Semantic Segmentation

Yongqin Xian*, Subhabrata Choudhury*, Yang He, Bernt Schiele, Zeynep Akata
CVPR 2019
pdfproject pagecodeBibTeX
Extenstion of zero-shot learning to segmentation: perform dense labelling of 'unseen' test classes by training on mutually exclusive 'seen' training classes with additional help from side information