What question did this study set out to answer?

To improve text-based person search by developing semantics-preserving feature representations through automated attribute mining and prediction.

March 1, 2020

Text-based Person Search via Attribute-aided Matching

View Full Paper

Ask AI

Bookmark

Authors

SASurbhi AggarwalIndian Institute of Science Bangalore RBR. Venkatesh BabuSri Venkateswara Veterinary University ACAnirban ChakrabortyNitte University

Discussion

Member takes

Implication

Algorithm evaluation demonstrates improved text-based person retrieval accuracy using mined attribute embeddings, highlighting the value of dual semantic and identity feature spaces.

Key Points

To improve text-based person search by developing semantics-preserving feature representations through automated attribute mining and prediction.
Extracted pedestrian descriptive attributes automatically from unstructured text descriptions to compensate for missing attribute annotations.
Constructed a dual-space retrieval framework combining an attribute-driven semantic space with an identity-driven feature space to bridge the image-text modality gap.
Evaluated matching and retrieval performance on the CUHK-PEDES benchmark dataset.
Achieved a state-of-the-art Rank-1 accuracy of 56.68% on the CUHK-PEDES dataset.
Generated cross-modal feature representations that are interpretable by human operators.

Cite This Study

Aggarwal et al. (2020) studied this question.

synapsesocial.com/papers/6a0fd57ad13714ec96feac4e https://doi.org/10.1109/wacv45572.2020.9093640

View Full Paper

Ask AI

Bookmark

Implication

Algorithm evaluation demonstrates improved text-based person retrieval accuracy using mined attribute embeddings, highlighting the value of dual semantic and identity feature spaces.

Text-based Person Search via Attribute-aided Matching

Discussion

Member takes

Implication

Key Points

Cite This Study

Text-based Person Search via Attribute-aided Matching

Discussion

Member takes

Implication

Key Points

Cite This Study

Also Consider

Also Consider