Vision, Learning, Imaging

One of our core areas of research activity is in computer vision, machine learning, and computational imaging. We have been fusing methodologies as diverse as Riemannian geometry, deep learning, compressive sensing etc. to develop general foundational techniques for processing of imaging data which excel under resource-constraints. Application areas include high-level scene analysis for consumer, defense, and mobile robotic platforms, where one is concerned with both accuracy requirements as well as operational constraints due to limitations in computational platforms. In this research thread, we focus on applications like activity, object, and scene analysis with special focus on developing techniques which are effective under impoverished measurement conditions. A sampling of our work in this area is given below.

This paper proposes a novel prior which is derived using basic theorems from probability theory and off-the-shelf optimizers, to improve fidelity of image generation using GANs by interpolating along any Euclidean straight line without any additional training and architecture modifications.
  • Rajhans Singh, Pavan Turaga, Suren Jayasuriya, Ravi Garg, Martin W. Braun, “Non-Parametric Priors For Generative Adversarial Networks”, in International Conference on Machine Learning (ICML), 2019. [pdf]

ReconNet (Kulkarni et al CVPR 2016, IEEE TCI 2018), a deep convolutional neural-net to recover high quality images from highly undersampled compressive measurements.
  • S. Lohit, K. Kulkarni, R. Kerviche, P. Turaga and A. Ashok, “Convolutional Neural Networks for Non-iterative Reconstruction of Compressively Sensed Images,” in IEEE Transactions on Computational Imaging. [pdf]
  • K. Kulkarni, S. Lohit, P. Turaga, R. Kerviche, A. Ashok, “ReconNet: Non-Iterative Reconstruction of Images from Compressively Sensed Random Measurements”, at the IEEE International Conference on Computer Vision and Pattern Recognition (CVPR), 2016. [pdf]

Reconstruction-free activity detection and activity recognition from challenging test cases, from compressively acquired imagery. One of the first works to demonstrate reconstruction-free compressive high-level visual inference on realistic data (Kulkarni et al PAMI 2016).
  • K. Kulkarni, P. Turaga, “Reconstruction-free action inference from compressive imagers”, at the IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), April 2016. [pdf]


Our algorithms for reconstruction-free inference have been tested with real data acquired from collaborating labs (Sankaranarayan, CMU) for the specific case of face-recognition (Lohit et al CCD 2015). This work won the best paper award at CCD 2015.
  • S. Lohit, K. Kulkarni, P. Turaga, J. Wang, A. Sankaranarayanan, “Reconstruction-free Inference on Compressive Measurements”, at the 4th IEEE Workshop on Computational Cameras and Displays (CCD), in conjunction with IEEE CVPR, June 2015. Best Paper Award [pdf]
  • Aswin Sankaranarayanan, Matthew Herman, Pavan Turaga, Kevin Kelly, “Enhanced Compressive Imaging Using Model-Based Acquisition: Smarter sampling by incorporating domain knowledge”, IEEE Signal Processing Magazine, Sep 2016. [pdf]
  • H. C. Braun, P. Turaga, A. S. Spanias, “Direct Tracking from Compressive Imagers: A Proof of Concept”, in IEEE International Conference Acoustics, Speech and Signal Processing (ICASSP), May 2014. [pdf]
  • A. Sankaranarayanan, P. Turaga, R. Chellappa, R. G. Baraniuk, “Compressive Acquisition of Linear Dynamical Systems”, SIAM Journal of Imaging Sciences, vol. 6, issue 4, 2013. [pdf]