On Symbiosis of Attribute Prediction and Semantic Segmentation

TL;DR

This paper introduces a novel approach combining semantic segmentation with attribute prediction to improve localization and accuracy of attribute detection in images, using weak supervision and a new symbiotic augmentation technique.

Contribution

The paper proposes Symbiotic Augmentation, a method that learns one mask per activation channel, enabling better attribute localization and prediction with reduced memory usage.

Findings

Outperforms previous methods on CelebA and LFWA datasets

Achieves superior attribute prediction accuracy

Enables attribute localization with weak supervision

Abstract

In this paper, we propose to employ semantic segmentation to improve person-related attribute prediction. The core idea lies in the fact that the probability of an attribute to appear in an image is far from being uniform in the spatial domain. We build our attribute prediction model jointly with a deep semantic segmentation network. This harnesses the localization cues learned by the semantic segmentation to guide the attention of the attribute prediction to the regions where different attributes naturally show up. Therefore, in addition to prediction, we are able to localize the attributes despite merely having access to image-level labels (weak supervision) during training. We first propose semantic segmentation-based pooling and gating, respectively denoted as SSP and SSG. In the former, the estimated segmentation masks are used to pool the final activations of the attribute prediction network, from multiple semantically homogeneous regions. In SSG, the same idea is applied to the intermediate layers of the network. SSP and SSG, while effective, impose heavy memory utilization since each channel of the activations is pooled/gated with all the semantic segmentation masks. To circumvent this, we propose Symbiotic Augmentation (SA), where we learn only one mask per activation channel. SA allows the model to either pick one, or combine (weighted superposition) multiple semantic maps, in order to generate the proper mask for each channel. SA simultaneously applies the same mechanism to the reverse problem by leveraging output logits of attribute prediction to guide the semantic segmentation task. We evaluate our proposed methods for facial attributes on CelebA and LFWA datasets, while benchmarking WIDER Attribute and Berkeley Attributes of People for whole body attributes. Our proposed methods achieve superior results compared to the previous works.