I am a research assistant professor at TTI Chicago, a philanthropically endowed academic computer science institute. I obtained my Ph.D. under the supervision of Jitendra Malik from the University of California at Berkeley, in 2011. In the past I have worked at Google, INRIA's LEAR group, Microsoft Research Bangalore, and the CLSP center at Johns Hopkins University.

I work in computer vision, machine learning and AI in general. I am interested in models for visual category recognition; in particular methods for localizing objects and estimating their properties from images; and studying the computational bottlenecks in building such systems.

• Ph.D. Dissertation | Research Statement | CV | Publications | Google Scholar

News & Updates

• One paper accepted for publication at CVPR 2013.
• Our paper on efficient evaluation of additive kernel SVMs appeared in the January 2013 edition of PAMI.
• Together with Nikhil Rasiwasia (Yahoo! Labs, Bangalore), we are organizing a tutorial at ICVGIP 2012 at IIT Bombay.
• Together with Jianxin Wu (Nanyang Technological University, Singapore), Andrea Vedaldi (University of Oxford, UK) and Florent Perronnin (Xerox Research Center Europe) we will give a half day tutorial on additive kernels and explicit embeddings. More details here.
• I have two workshop papers at ECCV 2012, one on learning a lexicon of attrbutes and parts, another on learing additive classifiers based on splines and Fourier representations (code available below).
• Together with Greg Shakhnarovich we presented a paper at HCOMP 2012 on collecting part annotations via pairwise correspondence.
• Together with Andrea Vedaldi, Iasonas Kokkkinos, Esa Rathu, Matthew Blaschko, Ross Girshick and Ben Taskar, I organized a six-week workshop at the JHU CLSP center on detailed image understanding.
• I have a TPAMI paper accepted for publication on additive kernels (with Alex Berg and Jitendra Malik).
• We have released several annotations on the PASCAL VOC dataset (see below).
• I have graduated from UC Berkeley and moved to TTI Chicago since January 2012.
• I spoke at the Mysore Park Workshop on Computer Vision in Dec 2011.
• I have two papers at ICCV 2011. One on predicting attributes ofpeople in the wild, and another on semantic contour detection. A large segmentation dataset containing all of PASCAL VOC 2011 detection images as well as the attributes dataset is now available for download.
• We submitted our CVPR 2011 action recognition algorithm to the PASCAL VOC 2011 action recognition challenge. Keypoint annotations for all VOC 2011 action trainval images is now available for download.
• Code & details for collecting various kinds of annotations on MTURK is now available.

Code, Datasets & Details

• C++ version of poselets available for UNIX and Windows. Mex file included. Significantly faster than the Matlab versions above. Includes models for all 20 PASCAL categories (from Lubomir Bourdev).
• Latest code for training additive classifiers with low memory overhead libspline-release1.0.tar.gz.
• Keypoint annotations and foreground masks for all objects in PASCAL VOC 2011 trainval images.
• Keypoint annotations for PASCAL VOC 2011 action images.
• Semantic boundaries dataset from our ICCV 2011 paper on semantic contour detection.
• Attributes of people dataset from our ICCV 2011 paper on attribute recognition.
• Efficient Additive Kernel SVM Classification:[Code Update] Constant time additive kernel SVM classification.
• Efficient Additive Kernel SVM Training: Linear time training and testing of additive classifiers.
• Mechanical Turk Tools: Large Scale Annotations on AMT.
• Poselets: [Code Update] Supervised parts for object detection, segmentation, pose estimation, etc.
• Action Classification: Static image action classification on PASCAL VOC 10.
• Pose Estimation: 3D human pose estimation from a single image.
• Biased Normalized Cuts: Efficient interactive normalized cuts.
• INRIA Pedestrian Detector: SPHOG + IKSVM based pedestrian detector.
• Max-Margin Hough Transform: Discriminative spatial models for parts.
• Handwritten Digit Classification on MNIST and USPS datasets using SPHOG + IKSVM classifiers.

Publications, Talks, etc.

Part Discovery from Partial Correspondence [pdf]
Subhransu Maji, Greg Shakhnarovich
CVPR 2013, Portland, Oregon (to appear)

Efficient Classification for Additive Kernel SVMs [pdf]
Subhransu Maji, Alexander C. Berg and Jitendra Malik
IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), Vol 35, No. 1, January 2013

Discovering a Lexicon of Parts and Attributes [pdf]|[slides]
Subhransu Maji, Second International Workshop on Parts and Attributes, ECCV 2012.

Linearized Smooth Additive Classifiers [pdf]|[slides]
Subhransu Maji, Workshop on Web-scale Vision and Social Media, ECCV 2012.
Code Available libspline-release1.0.tar.gz

Part Annotations via Pairwise Correspondence [pdf]
Subhransu Maji, Greg Shakhnarovich, 4th Workshop on Human Computation, AAAI 2012 (Oral)

Algorithms and Representations for Visual Recognition [url]|[pdf]
Subhransu Maji, Ph.D. dissertation, University of California at Berkeley, Dec. 2011.

Describing People: A Poselet-Based Approach to Attribute Classification [pdf]|dataset
Lubomir Bourdev, Subhransu Maji and Jitendra Malik
ICCV 2011, Barcelona, Spain (Oral)
Here is the video of the talk at ICCV 2011, and here are the slides.

Semantic Contours from Inverse Detectors [pdf]|[dataset]
Bharath Hariharan, Pablo Arbelaez, Lubomir Bourdev, Subhransu Maji and Jitendra Malik
ICCV 2011, Barcelona, Spain

Action Recognition from a Distributed Representation of Pose and Appearance [pdf]
Subhransu Maji, Lubomir Bourdev and Jitendra Malik
CVPR 2011, Colorado Springs, USA
Code & Details for: Pose estimation | Action classification

Biased Normalized Cuts [pdf] [Code]
Subhransu Maji, Nisheeth Vishnoi and Jitendra Malik
CVPR 2011, Colorado Springs, USA

Object Segmentation by Alignment of Poselet Activations to Image Contours [pdf]
Thomas Brox, Lubomir Bourdev, Subhransu Maji and Jitendra Malik
CVPR 2011, Colorado Springs, USA

Detecting People Using Mutually Consistent Poselet Activations [pdf] [Code]
Lubomir Bourdev, Subhransu Maji, Thomas Brox and Jitendra Malik
ECCV 2010, Crete, Greece

PASCAL VOC 2010
Static action recognition, object detection and segmentation based on poselets trained on 18 out of the 20 PASCAL VOC categories. For more details check out the PASCAL VOC 2010 workshop page and the POSELETS page.

Max-Margin Additive Classifiers for Detection [pdf] [PPT] [Code]
Subhransu Maji and Alexander C. Berg
ICCV 2009, Kyoto, Japan (Oral)

PASCAL VOC 2009
People detection and segmentation based on poselets. For more details check out the PASCAL VOC 2009 workshop page and the POSELETS page.

Multiple-View Object Recognition in Band-Limited Distributed Camera Networks [pdf] [PPT]
Allen Y. Yang, Subhransu Maji, C. Mario Christoudias, Trevor Darell, Jitendra Malik and S. Shankar Sastry
ICDSC 2009, Como, Italy (Oral)

Distributed compression and fusion of nonnegative sparse signals for multiple-view object recognition [pdf]
Allen Y. Yang, Subhransu Maji, Kirak Hong, Posu Yan and S. Shankar Sastry
International Conference on Information Fusion, 2009. (best paper award)

Object Detection Using a Max-Margin Hough Transform [pdf] [PPT]
Subhransu Maji and Jitendra Malik
CVPR 2009, Miami, Florida (Oral)

Classification using Intersection Kernel Support Vector Machines is Efficient [pdf] [Source Code]
Subhransu Maji, Alexander C. Berg and Jitendra Malik
CVPR 2008, Anchorage, Alaska

Fast Unsupervised Alignment of Video and Text for Indexing/Names and Faces [pdf]
Subhransu Maji and Ruzena Bajcsy
Multimedia Semantics Workshop, ACM Multimedia 2007

Confidence Based updation of Motion Conspicuity in Dynamic Scenes
Vivek Kumar Singh, Subhransu Maji and Amitabha Mukerjee
CRV 2006, Québec City, Canada

Other Reports

Large Scale Image Annotations on Amazon Mechanical Turk [ report]|[website]
Subhransu Maji
EECS Department, UCB, Tech. Rep. UCB/EECS-2011-79, July 2011

Fast and Accurate Digit Classification [pdf] [Code]
Subhransu Maji and Jitendra Malik
EECS Department, UCB, Tech. Rep. UCB/EECS-2009-159, Nov. 2009

Random Projection Features and Generalized Additive Models, Spring '08,
CS281B:Statistical Learning Theory(Peter Bartlett), Course Project.

Part of Image Tagging, Spring '08,
CS294-19:Statistical Natural Language Processing (Dan Klien)
Course Project with Patrik Sundberg.

Alignment of Video and Text for Search/Names and Faces, Spring '07,
CS281A:Statistical Learning Theory(Peter Bartlett), Course Project.

Automatic Photograph Enhancement using Graph Cuts, Spring '07,
CS270:Combinatorial Algorithms and Data Structures(Richard Karp)

Stability of Quality of Service, Fall '06,
CS262A, Course Project with Avinash Varadarajan.

Study of Feature Descriptors for recognition, Fall '06,
CS294-6: Theoretical, Conceptual and Experimental Vision(bajcsy, sastry, yang).

Graph Classification: Applications in ChemoInformatics,
Undergraduate Thesis, Fall'06, Indian Institute of Technology, Kanpur.

Personal

I love to hike in my spare time. Here is me at 16,500 ft in the Himlayas -- a hike to the mystery lake at Roopkund. I also like to ski and bike around. I occasionally draw and paint. Here are some of my paintings. Back during my undergrad I used to create computer animations, some of which are here.