Krish Singh
I am a PhD student at Visual Inference Lab at TU Darmstadt with Dr. Stefan Roth , where I work on computer vision and machine learning.
I am interested in generative modeling with a particular focus on compositionality and disentanglement of scenes.
I previously spent three beautiful years at Rakuten research working on object detection, retrieval, and complementary item recommendation for fashion images.
I completed my graduation from IIT Hyderabad , where I was advised by Vineeth N Balasubramanian .
Email /
Bio /
Google Scholar /
Twitter /
Github
|
|
Research
I'm interested in computer vision, machine learning, optimization, and image processing.
Much of my research is focused on generative modeling with a focus on compositionality and disentanglement for images.
|
|
S2-Flow: Joint Semantic and Style Editing of Facial Images
Krishnakant Singh,
Simone Schaub-Meyer,
Stefan Roth,
BMVC, 2022
Code
/
video
/
arXiv
Disentangling the joint space of a pretrained GAN into semantic and style space using normalizing flow enables more precise edits in both the semantic and style domains.
|
|
Submodular Batch Selection for Training Deep Neural Networks
K J Joseph,
Vamshi Teja Racha,
Krishnakant Singh,
Vineeth N. Balasubramanian,
IJCAI, 2019 and ICML Workshop, 2019
Code
/
arXiv
/
Video
A Submodular objective for batch selection for creating diverse batches for training neural networks faster and more efficiently.
|
|
Submodular Importance Sampling for training Neural Networks
Krishnakant Singh
Masters's Thesis
Code
/
arXiv
An empirical and theoretical analysis of submodular functions for creating mini-batches for SGD. The thesis showed that our novel submodular objective empirically reduces the variance of SGD-type algorithms, leading to faster convergence of the optimization process.
|
|
Google Summer of Code with MLPack |
Summer Intern |
(May, 2017- Sep, 2017) |
|
Rakuten Research |
Research Scientist |
(Jan, 2020 - March, 2021) |
|
Rakuten Inc. |
Research Engineer |
(Oct, 2018- Jan, 2020) |
Website Credits to Mr. Jon Barron source code
Link to his website : Go
|
|