Object recognition and pose estimation

This page presents code and data related to the following paper:
Multi-view Feature Distributions for Object Detection and Continuous Pose Estimation,
Damien Teney, Justus Piater, Computer Vision and Image Understanding, 2014.


The following videos demonstrate the generative multiview model, which interpolates the appearance of an object at arbitrary viewpoints, being trained only on a limited set of discrete viewpoints (the images below). The dataset used is borrowed from Viksten et al.

Code and data

Download the Matlab code (includes a GUI for visualizing datasets and results).

Download a sample dataset, excerpt from S. Savarese's "3D Object" dataset, with our (learned) model of a car, and our (precomputed) results on test images.