Enhanced composed fashion image retrieval with a multi-hop reasoning framework

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Existing Composed Fashion Image Retrieval (CFIR) methods struggle with complex multimodal queries.
• Single-step and hierarchical retrieval methods fail to capture intricate visual-textual interplay in fashion datasets.

Purpose of the Study:

• To introduce a novel Multi-Hop Reasoning Framework for Composed Fashion Image Retrieval (CFIR).
• To enhance the interpretation of complex fashion queries by integrating visual and textual data more effectively.

Main Methods:

• Developed a Multi-Hop Reasoning Framework that iteratively refines retrieval processes.
• Employed multi-hop reasoning to manage compositionality in fashion-related queries.
• Integrated visual and textual data through multiple reasoning steps.

Main Results:

• Achieved superior retrieval accuracy and performance in Composed Fashion Image Retrieval.
• Demonstrated marked improvements over state-of-the-art methods on Fashion-IQ, Shoes, and Fashion200k datasets.
• Showcased enhanced ability to discern subtle relationships between images based on modification descriptions.

Conclusions:

• The Multi-Hop Reasoning Framework offers a significant advancement for Composed Fashion Image Retrieval.
• The proposed methodology sets a new benchmark in the field of image retrieval for complex fashion queries.
• Multi-hop reasoning effectively addresses the limitations of traditional CFIR approaches.