Attribute Aware Pooling for Pedestrian Attribute Recognition

Abstract

This paper expands the strength of deep convolutional neural networks (CNNs) to the pedestrian attribute recognition problem by devising a novel attribute aware pooling algorithm. Existing vanilla CNNs cannot be straightforwardly applied to handle multi-attribute data because of the larger label space as well as the attribute entanglement and correlations. We tackle these challenges that hampers the development of CNNs for multi-attribute classification by fully exploiting the correlation between different attributes. The multi-branch architecture is adopted for fucusing on attributes at different regions. Besides the prediction based on each branch itself, context information of each branch are employed for decision as well. The attribute aware pooling is developed to integrate both kinds of information. Therefore, attributes which are indistinct or tangled with others can be accurately recognized by exploiting the context information. Experiments on benchmark datasets demonstrate that the proposed pooling method appropriately explores and exploits the correlations between attributes for the pedestrian attribute recognition. * We thank anonymous reviewers for their helpful comments.

Beyond Human Parts: Dual Part-Aligned Representations for Person Re-Identification

“…Attributes for ReID. Semantic attributes [46,25,7] have been exploited as feature representations for person reidentification tasks. Previous work [47,6,20,42,58] leverages the attribute labels provided by original dataset to generate attribute-aware feature representation.…”

A multi-branch separable convolution neural network for pedestrian attribute recognition

“…The RA model makes use of the global spatial locality and local attention correlation to improve the overall robustness. Correlation between attributes is explored by [39]. Their multi-branch network collects context information to compute attribute probabilities.…”

Human Attribute Recognition: A Comprehensive Survey

“…In [59], authors discussed that a plain CNN could not handle human multi-attribute classifications effectively, as for each image, several labels have been entangled. To address this challenge, Han et al [59] proposed to use a ResNet50 backbone followed by multiple branches to predict the occurrence probability of each attribute. Further, to improve the results, they provided a matrix from ground truth labels to obtain the conditional probability of each label (semantic attribute) given another attribute.…”