Neural Network Image Retrieval for Body Type and Pose
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image retrieval systems lack the intelligence to identify and disentangle contextual features such as body type and pose of models in images, making it impossible for users to search for specific product-model pairings effectively, especially in online shopping where users struggle to visualize how clothing would look on their body.
Innovation Solution
The use of neural networks trained to classify body type and pose in images, allowing users to input specific criteria and retrieve images of models wearing articles of clothing or accessories with matching body types and poses, leveraging user-generated content and image processing to label and cluster images based on these features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If neural networks are used to classify body type and pose in images, then image retrieval precision is improved, but device complexity increases
Solution Approach 1:
The system segments the image analysis task into two separate neural networks: one dedicated to body type classification and another to pose classification. This segmentation allows each network to specialize in a specific feature, improving overall classification precision while making the system more manageable and less complex than a single monolithic network would be.
2Adaptability or versatility
If multiple neural networks are deployed for body type and pose classification, then adaptability is improved, but device complexity increases
Solution Approach 1:
The neural networks are designed with universal applicability - they can classify multiple body types and poses without requiring separate models for each category. The networks are trained on diverse datasets encompassing various body types (e.g., athletic, average, curvy) and poses (front, side, back views), enabling the system to adapt to different classification needs using the same infrastructure.
3Measurement precision
If images are labeled and clustered based on body type and pose, then image retrieval precision is improved, but loss of time increases
Solution Approach 1:
The system performs preliminary labeling and clustering of images by body type and pose during the image ingestion phase, before actual retrieval operations occur. This pre-processing creates organized clusters that can be quickly queried during user searches, trading initial processing time for significantly faster retrieval performance when users need to find specific images.
Data Source
AI summary
Techniques are provided for providing a user with retrieved images of a specific article of clothing or accessory worn by models having a particular body type and pose as selected by the user. The images have been analyzed to identify both a body type and pose of the model wearing the article of clothing in each of the images. The images are labeled based on at least body type and pose of the model and clustered based on their labels, and are thus available for retrieval by subsequent requests by a user. In particular, a user that is interested in the article of clothing can input one or more requests for images of models having a selected body type and pose. Any of the images labeled with the selected body type and pose can then be provided to the user in any number ways.