Aayush Bansal

I am a PhD student (2015-) at the Robotics Institute of Carnegie Mellon University, where I am supervised by Deva Ramanan and Yaser Sheikh. I received my Masters (2013-15) under Abhinav Gupta. At CMU, I have also worked with Daniel Huber on real-time systems and robotic applications. Prior to joining CMU as a graduate student, I worked with Devi Parikh. I did my undergrad in ECE from University of Delhi in 2011.

I had a joyful summer (2017) working with the amazing Shugao Ma, and my advisor, Yaser Sheikh, at Oculus Research in Pittsburgh. Previously, I spent a wonderful summer (2015) working with Bryan Russell at Adobe Research in San Francisco.

I received the prestigious Qualcomm Fellowship for 2017-18, an Uber Presidential Fellowship for 2016-17, and named a Presidential Fellow at CMU.

Email: aayushb [at] cs [dot] cmu [dot] edu



Do Convolutional Neural Networks act as Compositional Nearest Neighbors?
Chunhui Liu, Aayush Bansal, Victor Fragoso, Deva Ramanan


PixelNN: Example-based Image Synthesis
Aayush Bansal, Yaser Sheikh, Deva Ramanan
International Conference on Learning Representations (ICLR), 2018
project page / arXiv

Simple pixel-wise nearest neighbors at work!


PixelNet: Representation of the pixels, by the pixels, and for the pixels.
Aayush Bansal, Xinlei Chen, Bryan Russell, Abhinav Gupta, Deva Ramanan
Under review for Transactions on Pattern Analysis & Machine Intelligence (TPAMI)
project page / arXiv / codes


Marr Revisited: 2D-3D Alignment via Surface Normal Prediction
Aayush Bansal, Bryan Russell, Abhinav Gupta
Computer Vision and Pattern Recognition (CVPR), 2016
project page /arXiv preprint / codes


Mid-level Elements for Object Detection
Aayush Bansal, Abhinav Shrivastava, Carl Doersch, Abhinav Gupta
arXiv / analysis

We proposed an object detection approach using HoG-based mid-level elements. Without using any contextual reasoning and external data, this method achieved a mean average precision (mAP) of 41.9 on PASCAL VOC 2007, and mAP of 37.1 on VOC 2010. This is comparable to existing state-of-the-art techniques on comp-3 evaluation.


Towards Transparent Systems: Semantic Characterization of Failure Modes
Aayush Bansal, Ali Farhadi, Devi Parikh
European Conference on Computer Vision (ECCV), 2014
project page / supplementary material

We proposed an approach that identifies patterns in failures and summarizes them with a human understandable characterization using attributes.


Understanding How Camera Configuration and Environmental Conditions Affect Appearance-based Localization
Aayush Bansal, Hernan Badino, Daniel Huber
IEEE Intelligent Vehicles (IV), 2014
project page

We analyzed different sensor configurations and environmental conditions for topometric localization algorithm.


Which Edges Matter?
Aayush Bansal, Adarsh Kowdle, Devi Parikh, Andrew Gallagher, Larry Zitnick
Workshop on 3D Representation and Recognition (3dRR) at ICCV, 2013
project page / presentation

We evaluated the information conveyed by different edges through human studies.


CANINE : A robotic mine dog
B. A. Stancil, J. Hyams, J. Shelly, K. Babu, H. Badino, A. Bansal, D. Huber, P. Batavia
IS&T Conference on Electronic Imaging (SPIE), 2013
project page / video / competition rules

I wrote the object detection module for this robot.


Geometry-based Methods in Computer Vision (16-822), CMU
Teaching Assistant (TA) with Prof Martial Hebert
Fall 2017

Computer Vision (16-720), CMU
Teaching Assistant (TA) with Prof Srinivasa Narasimhan
Spring 2015

Thanks to Jon Barron for the webpage design!