Appearance-based object detection systems using statistical models have proven quite successful. They can reliably detect textured, rigid objects in a variety of poses, lighting conditions and scales. However, the construction of these systems is time-consuming and difficult because a large number of training examples must be collected and manually labeled in order to capture variations in object appearance. Typically, this requires indicating which regions of the image correspond to the object to be detected, and which belong to background clutter, as well as marking key landmark locations on the object. The goal of this work is to pursue and evaluate approaches which reduce the amount of fully labeled examples needed, by training these models in a semi-supervised manner. To this end, we develop approaches based on Expectation-Maximization and self-training that utilize a small number of fully labeled training examples in combination with a set of weakly labeled examples. This is advantageous in that weakly labeled data are inherently less costly to generate, since the label information is specified in an uncertain or incomplete fashion. For example, a weakly labeled image might be labeled as containing the training object, with the object location and scale left unspecified. In this work we analyze the performance of the techniques developed through a comprehensive empirical investigation. We find that supplementing a small fully labeled training set with weakly labeled data in the training process reliably improves detector performance for a variety of detection approaches. The outcome is the identification of successful approaches and key issues that are central to achieving good performance in the semi-supervised training of object detection systems.

138 pages

Return to: SCS Technical Report Collection
School of Computer Science homepage

This page maintained by reports@cs.cmu.edu