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

This page is intended to provide additional details that could not be accomodated in the following paper due to space constraints and provide an access mechanism to the data set used in this paper. In future, we also intend to share the code for the model proposed in our paper:

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.

FlickrSports Data set

We collected sports images from Flickr, an online photo management and sharing application that provides an API that supports multiple word queries for searching, listing, and downloading images. We used some sports team names and venues as queries to construct a data set of images of five popular sports: baseball, basketball, football, soccer, and tennis. We discarded the images without a significant view of the playing field, but did not restrict the images to include the entire view of the field. Some of the images include players, balls, or other objects, occluding the distinctive markings on the ground. The data set contains 2449 images with roughly the same number of images for each sport. We split the data set into three parts: 50\% for training, 25\% for validation, and 25\% for testing. The training and validation sets are used for tuning the parameters, and the test set is used for the final evaluation. While we are working on the issues related to usage permissions and standardization of data splits, the data set is available for non-commercial, academic use only by contacting Vidit Jain.