Papers and reports by theme | type

My research follows these main themes:

• Recognition architectures for object detection, attribute recognition, pose estimation, fine-grained recognition, materials/texture recognition, etc.
• Grouping and segmentation methods.
• Human computation for collecting annotations, recognition with humans "in the loop".
• Statistical tools for learning and inference from data.

Recognition

Deep Filter Banks for Texture Recognition and Segmentation Mircea Cimpoi, Subhransu Maji, and Andrea Vedaldi CVPR 2015, Boston (Oral) abstract A state of the art texture descriptor achieving an accuracy of 79.8% on Flickr material dataset and 81% on MIT indoor scenes dataset.

Automatic Image Annotation using Deep Learning Representations Venkatesh N. Murthy, Subhransu Maji, and R. Manmatha. ICMR 2015, Shanghai, China (short paper)

Knowing a Good HOG Filter When You See It: Efficient Selection of Filters for Detection Ejaz Ahmed, Gregory Shakhnarovich, and Subhransu Maji ECCV 2014, Zurich (Oral) webpage (coming soon) / talk slides / poster

Understanding Objects in Detail with Fine-grained Attributes A. Vedaldi, S. Mahendran, S. Tsogkas, S. Maji, B. Girshick, J. Kannala, E. Rahtu, I. Kokkinos, M. B. Blaschko, D. Weiss, B. Taskar, K. Simonyan, N. Saphra, and S. Mohamed CVPR 2014, Columbus, Ohio webpage / spotlight / poster

Describing Textures in the Wild Mircea Cimpoi, Subhransu Maji, Iasonas Kokkinos, Sammy Mohamed, and Andrea Vedaldi CVPR 2014, Columbus, Ohio webpage / dataset / spotlight / poster

Part Discovery from Partial Correspondence Subhransu Maji and Gregory Shakhnarovich CVPR 2013, Portland, Oregon poster Discover semantically meaningful parts from correspondence annotations.

Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev, Subhransu Maji and Jitendra Malik ICCV 2011, Barcelona, Spain (Oral) attribute annotations / talk slides

Action Recognition from a Distributed Representation of Pose and Appearance Subhransu Maji, Lubomir Bourdev and Jitendra Malik CVPR 2011, Colorado Springs, USA poster / pose estimation webpage / action classification webpage Models for predicting the action and 3d pose of humans from still images.

Detecting People Using Mutually Consistent Poselet Activations Lubomir Bourdev, Subhransu Maji, Thomas Brox and Jitendra Malik ECCV 2010, Crete, Greece webpage / poster A top-performing person detector on the PASCAL VOC dataset.

Multiple-View Object Recognition in Band-Limited Distributed Camera Networks Allen Y. Yang, Subhransu Maji, Mario C. Christoudias, T. Darrell, J. Malik, and S. S. Sastry ICDSC 2009, Como, Italy (Oral) talk slides

Object Detection Using a Max-Margin Hough Transform Subhransu Maji and Jitendra Malik CVPR 2009, Miami, Florida (Oral) talk slides

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

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

Fine-Grained Visual Classification of Aircraft Subhransu Maji, Esa Rahtu, Juho Kannala, Matthew Blaschko and Andrea Vedaldi This dataset will be a part of the Fine-Grained Visual Recognition 2013 Challenge FGVC workshop poster / webpage

Fast and Accurate Digit Classification Subhransu Maji and Jitendra Malik, EECS Dept., UCB, Tech. Rep. UCB/EECS-2009-159, Nov. 2009 webpage An analysis of histogram based features and non-linear classifiers for digit recognition leading to a strong baseline that achieves 0.79% error on the MNIST dataset and 3.4% error on the USPS dataset.

Grouping and segmentation

Parsing World's Skylines using Shape Constrained MRFs Rashmi Tonge, Subhransu Maji, and C.V. Jawahar CVPR 2014, Columbus, Ohio webpage / code / dataset / spotlight / poster

Semantic Contours from Inverse Detectors Bharath Hariharan, Pablo Arbelaez, Lubomir Bourdev, Subhransu Maji and Jitendra Malik ICCV 2011, Barcelona, Spain semantic contour dataset Contains boundary annotations for all objects in the PASCAL VOC 2011 dataset.

Biased Normalized Cuts Subhransu Maji, Nisheeth Vishnoi and Jitendra Malik CVPR 2011, Colorado Springs, USA webpage / poster Linear time normalized cuts with constraints.

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

Human computation

Learning Localized Perceptual Similarity Metrics for Interactive Categorization Catherine Wah, Subhransu Maji, and Serge Belongie WACV, Waikoloa Beach, HI, 2015. Best paper award

Part and Attribute Discovery from Relative Annotations Subhransu Maji and Gregory Shakhnarovich International Journal of Computer Vision May 2014, Volume 108, Issue 1-2, pp 82-96

Similarity Comparisons for Interactive Fine-Grained Categorization Catherine Wah, Grant V. Horn, Steven Branson, Subhransu Maji, Pietro Perona, and Serge Belongie CVPR 2014, Columbus, Ohio spotlight / poster

Discovering a Lexicon of Parts and Attributes Subhransu Maji, Second International Workshop on Parts and Attributes, ECCV 2012. talk slides / fgvc workshop poster Crowdsourced discovery of discriminative attributes. Best poster runner-up FGVC Workshop, CVPR 2013

Part Annotations via Pairwise Correspondence Subhransu Maji and Gregory Shakhnarovich 4th Workshop on Human Computation, AAAI 2012 (Oral) talk slides

Large Scale Image Annotations on Amazon Mechanical Turk Subhransu Maji, EECS Dept., UCB, Tech. Rep. UCB/EECS-2011-79, July 2011 webpage Tools for collecting image annotations via crowdsourcing.

Machine learning

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

Active Boundary Annotation using Random MAP Perturbations Subhransu Maji, Tamir Hazan and Tommi Jaakkola AISTATS 2014, Reykjavik, Iceland poster A framework for coarse-to-fine boundary annotation.

Learning Efficient Random Maximum A-Posteriori Predictors with Non-Decomposable Loss Functions Tamir Hazan, Subhransu Maji, Joseph Keshet and Tommi Jaakkola NIPS 2013, Lake Tahoe, Nevada poster

On Sampling from the Gibbs Distribution with Random Maximum A-Posteriori Perturbations Tamir Hazan, Subhransu Maji and Tommi Jaakkola NIPS 2013, Lake Tahoe, Nevada poster

Linearized Smooth Additive Classifiers Subhransu Maji, Workshop on Web-scale Vision and Social Media, ECCV 2012. code / talk slides This paper looks at training additive classifiers from the perspective of "generalized additive models".

Max-Margin Additive Classifiers for Detection Subhransu Maji and Alexander C. Berg ICCV 2009, Kyoto, Japan (Oral) talk slides / website / code (pwl_sgd.tar.gz) Train additive classifiers an order-of-magnitude faster with no loss in accuracy.

Distributed Compression and Fusion of Non-negative Sparse Signals for Multiple-view Object Recognition Allen Y. Yang, Subhransu Maji, Kirak Hong, Posu Yan, Shankar S. Sastry International Conference on Information Fusion, 2009. Best paper award

Classification using Intersection Kernel Support Vector Machines is Efficient Subhransu Maji, Alexander C. Berg and Jitendra Malik CVPR 2008, Anchorage, Alaska webpage Evaluate non-linear kernel SVMs orders of magnitude faster!