Vidit Jain PhD candidate Computer Vision Laboratory Computer Science Department University of Massachusetts Amherst vidit -AT- cs -DOT- umass -DOT- edu Office: Room 256, #1 Ph. 413-545-0528

Research Interests

To follow the above-mentioned approach (for vision tasks), I am interested in exploring probabilistic graphical models (both generative and discriminative) and geometric methods (manifold learning). In general, I am interested in identifying the characteristics of the problem and the plausible models for their applicability (as opposed to subscribing to a particular form of solutions).

I organize my research notes and general observations about computer vision and machine learning on my blog titled "Learning in Vision".

At Umass, I have been involved in the following projects:

- Hyper-features

- Face Recognition

- Face Alignment

- MR bias correction

Affiliations

1998-2002	2001	2002-04	2004-	2006	2007	2009
[College] Manindra Agrawal A. Mukerjee (Discrete Math, AI)	[Intern] Jop Sibeyn (Parallel Algo)	[Technical Lead] Thomas O. Binford (Handwriting Recognition)	[Grad School] E. Learned-Miller Andrew McCallum Allen Hanson (Vision, ML)	[Intern] M. Narasimhan Paul Viola (Information Extraction)	[Intern] A. Singhal Jiebo Luo (Event Classificaton)	[Intern] M. Varma P. Anandan (Ranking for Image Search)

Publications

• Vidit Jain, Amit Singhal, Jiebo Luo.

  Selective Hidden Random Fields: Exploiting Domain Specific Saliency for Event Classification.

  In Proceedings of IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2008, Anchorage, Alaska, USA.

  (Hidden-state conditional random fields for a joint modeling of segmentation and classification. It provides a principled approach for selecting features - or image regions - suitable for classification. Experiments are performed for a very difficult sports classification task.)
• Vidit Jain, Amit Singhal, Jiebo Luo, Erik Learned-Miller.

  Exploiting Domain Specific Saliency for Event Classification.

  Abstract and Poster Presentation at the Scene Understanding Symposium (SUnS) 2008, Boston MA USA.

  (See our CVPR'08 paper for an extended version)
• Vidit Jain, Erik Learned-Miller, Andrew McCallum.

  People-LDA: Anchoring Topics to People using Face Recognition.

  In Proceedings of International Conference on Computer Vision (ICCV), 2007, Rio De Janeiro, Brazil.

  (Jointly model people's identity, face appearance in an image, and surrounding text in the image captions with an LDA-style topic model. Improved results in identifying coherent sets of person "mentions"---that is, improved co-reference by using both text and image features.)
• Gary B. Huang, Vidit Jain, Erik Learned-Miller.

  Unsupervised Joint Alignment of Complex Images.

  In Proceedings of International Conference on Computer Vision (ICCV), 2007, Rio De Janeiro, Brazil.

  (Automatically build an alignment mechanism for complex classes, such as human faces and cars, without any labeling of parts or poses.)
• Vidit Jain, Andras Ferencz, Erik Learned-Miller.

  Discriminative Training of Hyper-feature Models for Object Identification.

  In Proceedings of British Machine Vision Conference (BMVC), 2006, Edinburgh UK.

  (The differences in appearances of pairs of objects are modeled to obtain effective class-specific identifiers. A direct optimization to maximize the conditional likelihood of the identification label as "same" or "different" given the observation results in the state-of-the-art face identifier.)
• Vidit Jain, Allen Hanson, Erik Learned-Miller, Edward Riseman, Joseph Horowitz, Benjamin Liptzin.

  A New Approach to Magnetic Resonance Imaging for Early Diagnosis of Alzheimer's Disease.

  Internal Report, Collaborative Biomedical Research Program, UMass Amherst and Baystate Health Systems, 2005

  (A literature review of recent advances in the diagnosis of Alzheimer's Disease and a proposal for atrophy measurements for an early diagnosis.)
• Erik Learned-Miller and Vidit Jain.

  Many heads are better than one: Jointly removing bias from multiple MRIs using nonparameteric maximum likelihood.

  In Proceedings of Information Processing in Medical Imaging (IPMI), 2005, Glenwood Spring CO USA.

  (Joint alignment of three-dimensional magenitic resonance imaging volumes of infant brains.)
• Vidit Jain, Marghoob Mohiyuddin and Phalguni Gupta.

  Parallel Bi-dimensional Pattern Matching with Scaling.

  In Proceedings of IASTED Conference on Parallel and Distributed Systems (PDCS), 2002, Boston USA.

  (Parallelized version of Amir et al.'s algorithm for two-dimensional pattern matching.)
• Vidit Jain.

  Analysis of Pseudorandom Permutations.

  Undergraduate thesis, Dept. of CSE, IIT Kanpur, 2002.

  (Proposes a pseudo-random permutation generator. Three statistical tests are designed to evaluate the pseudo-randomness of a permutation generator. The results show that DES, when used as an oracle for finding the successor of a number with the seed generated by a pseudo-random number generator, can be used as a pseudo-random permutation generator.)
• Vidit Jain, Jop Frederik Sibeyn.

  Determination of Cycle Structure in Large Permutations : Parallel Implementation on a Cluster of Workstations.

  Technical Report, Department of Computing Science, Umea University, 2001, Umea Sweden.