Zhang Ning Paluri Manohar Ranzato Marc'Aurelio Darrell Trevor Bourdev Lubomir

Abstract

We propose a method for inferring human attributes (such as gender, hairstyle, clothes style, expression, action) from images of people under largevariation of viewpoint, pose, appearance, articulation and occlusion.Convolutional Neural Nets (CNN) have been shown to perform very well on largescale object recognition problems. In the context of attribute classification,however, the signal is often subtle and it may cover only a small part of theimage, while the image is dominated by the effects of pose and viewpoint.Discounting for pose variation would require training on very large labeleddatasets which are not presently available. Part-based models, such as poseletsand DPM have been shown to perform well for this problem but they are limitedby shallow low-level features. We propose a new method which combinespart-based models and deep learning by training pose-normalized CNNs. We showsubstantial improvement vs. state-of-the-art methods on challenging attributeclassification tasks in unconstrained settings. Experiments confirm that ourmethod outperforms both the best part-based methods on this problem andconventional CNNs trained on the full bounding box of the person.