An empirical study of visual features for part based model

Object detection is a fundamental task in computer vision. Deformable part based model has achieved great success in the past several years, demonstrating very promising performance. Many papers emerge on part based model such as structure learning, learning more discriminative features. To help researchers better understand the existing visual features' potential for part based object detection and promote the deep research into part based object representation, we propose an evaluation framework to compare various visual features' performance for part based model. The evaluation is conducted on challenging PASCAL VOC2007 dataset which is widely recognized as a benchmark database. We adopt Average Precision (AP) score to measure each detector's performance. Finally, the full evaluation results are present and discussed.