2022-Complex Intelligent Systems Survey on clothing image retrieval with cross-domain

Motivation

A survey aiming at achieving accurate clothing-retrieval in cross-domain situations.

Survey Results

Challenges

• clothes are flexible items, and when viewed from different shooting angles or The appearance can be very different when wearing different body types.

• The intra-class variance is large and the inter-class variance is
  small, which is an inherent characteristic of clothing images.

Previous work

1. critical region recognition

• Bounding box
• Humanbody landmark
• Clothing landmark
• Attention map

1. Deep metric learning

• Siamese network
• Triplet network and variants
• Ensemble network

Clothing databases

• Street2Shop
• DARN
• DeepFashion v1&v2 (Consumer-shop)
• ModaNet

Summary

• clothing landmark recognition and attention map recognition have a higher accuracy in clothing retrieval.
• deep metric learning is better in the same domain clothing image retrieval than the crossdomain clothing image retrieval.
• It can be seen that the overall effect of deep metric learning is not as good as clothing critical region recognition of solving cross-domain retrieval problems, indicating that the main problem to be solved in cross-domain clothing
  retrieval are the recognition of clothing important regions in the image.