Subhajit Chaudhury

I am a researcher at IBM Research AI, Tokyo where I work on reinforcement learning algorithms for vision-based robotic control and imitation learning. Concurrently, I am a first-year Ph.D. student at the University of Tokyo. Previously, I was a researcher at NEC Central Research Labs working on predictive infrastructure maintenance using deep learning.

I received my master's degree in Electrical Engineering from IIT Bombay in 2014, where I worked with Prof. Subhasis Chaudhuri in the field of haptics and human-computer interaction. I did my bachelor's in Electrical Engineering from Jadavpur University in 2012.

Email  /  CV  /  Google Scholar  /  LinkedIn


My research interests are in the intersection of reinforcement learning and computer vision. At IBM, I am working on robust reward estimation for imitation learning from videos. In my Ph.D. work, I am focusing on exploring interpretability in neural networks for computer vision.

Video Imitation GAN: Learning control policies by imitating raw videos using generative adversarial reward estimation
Subhajit Chaudhury, Daiki Kimura, Tu-Hoa Pham, Asim Munawar and Ryuki Tachibana
arXiv preprint arXiv:1810.01108., 2018
arxiv / video

Internal Model from Observations for Reward Shaping
Daiki Kimura, Subhajit Chaudhury, Ryuki Tachibana and Sakyasingha Dasgupta
International Conference of Machine Learning (ICML), Adaptive Learning Agents Workshop, 2018
arxiv / bibtex

Focusing on What is Relevant: Time-Series Learning and Understanding using Attention
Phongtharin Vinayavekhin, Subhajit Chaudhury, Asim Munawar, Don Joven Agravante, Giovanni De Magistris, Daiki Kimura and Ryuki Tachibana
International Conference on Pattern Recognition (ICPR), 2018
arxiv / bibtex

Transfer learning from synthetic to real images using variational auto-encoders for robotic applications
Tadanobu Inoue, Subhajit Chaudhury, Giovanni De Magistris and Sakyasingha Dasgupta
IEEE International Conference on Image Processing (ICIP), 2018
arxiv / Video / bibtex

Conditional generation of multi-modal data using constrained embedding space mapping International Conference on Machine Learning (ICML), Implicit Models Workshop, 2017
Text to image generative model using constrained embedding space mapping IEEE International workshop on Machine learning for Signal Processing (MLSP), 2017

Subhajit Chaudhury, Sakyasingha Dasgupta, Asim Munawar, Md. A. Salam Khan and Ryuki Tachibana
arxiv / Poster / Slides / bibtex

Can fully convolutional networks perform well for general image restoration problems?
Subhajit Chaudhury and Hiya Roy
International Conference on Machine Vision Applications (MVA), 2017
arxiv / Poster / bibtex

Spatial-Temporal Motion Field Analysis for Crack Detection on Concrete Surfaces
Subhajit Chaudhury, Gaku Nakano, Jun Takada, Akihiko Iketani
IEEE Winter Conference on Applications of Computer Vision (WACV), 2017
Paper / Poster / bibtex

Convolutional Neural Network Layer Re-ordering for Acceleration
Vijay Daultani, Subhajit Chaudhury, Kazuhisa Ishizaka
Workshop on Synthesis And System Integration of Mixed Information (SASIMI), 2016
Paper / Poster / bibtex

Volume Preserving Haptic Pottery (Oral Presentation)
Subhajit Chaudhury and Subhasis Chaudhuri
IEEE Haptics Symposium (HAPTICS), 2014
Paper / Video / bibtex

Vision-Based Human Pose Estimation for Virtual Cloth Fitting.
Sourav Saha, Pritha Ganguly, and Subhajit Chaudhury
Indian Conference on Computer Vision, ICVGIP, 2014
Paper / bibtex

Other Projects
Classification of Hyperspectral Satellite Images Using Deep Convolutional Neural Networks
Subhajit Chaudhury, Hiya Roy

Implemented land-cover classification from hyperspectral satellite images using convolutional neural networks. Results demonstrate that CNN based classification performs considerably well compared to traditional feature classifiers like support vector machines.

Vanilla ray-tracing algorithm
Satwik Kottur, Subhajit Chaudhury

Implemented vanilla ray-tracing from scratch given a set of objects in the scene, light source, and surface reflection parameters.

The Music Box : Computer Graphics animation in OpenGL from scratch.
Subhajit Chaudhury, Urmi Sadhu

Modeled an articulated human figure using basic geometry transformations, prepared a virtual environment, provided appropriate lighting and texture maps to finally create a full animation in OpenGL from scratch.

Estimating disparity map from stereo images
Subhajit Chaudhury, Avik Hati

Obtained the pixel-wise disparity map from pixel correspondence using block-wise correlation matching.

Shape from Shading Using Surface Normals
Avik Hati, Subhajit Chaudhury

Estimated the shape of an object using the intensity value from the image. We used the classical "shape from shading" using p, q values, and global minimization technique.


Teaching Assistant, EE210 Signals and System, IIT Bombay (Spring-2013)

Teaching Assistant, EE603 Digital Signal Processing and its application, IIT Bombay (Autumn-2013)

Teaching Assistant, Computer Vision, IIT Bombay (Spring 2014)

Original Website Credits