Deep Learning Virtual Image Generation for Product Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Online shopping malls face challenges in effectively displaying product images, as existing images are often displayed randomly, leading to poor product recognition and the need for separate photographing or retouching to create additional images, which is inconvenient and costly.
Innovation Solution
A deep-learning based method that classifies and creates virtual product images by determining target categories and using pre-trained neural networks to generate images from existing product images, eliminating the need for separate photographing or retouching.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If product images are displayed randomly on the shopping mall page, then the system complexity is reduced, but product recognition effectiveness deteriorates
Solution Approach 1:
The patent uses deep learning models to generate virtual images that copy and transform existing product images into different pose categories. Instead of requiring multiple physical photographs, the system creates synthetic images through image synthesis techniques, maintaining product recognition effectiveness while reducing the need for complex image acquisition systems
Solution Approach 2:
The patent replaces the mechanical process of separate photographing and retouching work with an automated deep learning-based image generation system. The neural network models automatically transform existing images into different poses, eliminating the need for manual photography operations and reducing system complexity
2Adaptability or versatility
If separate photographing or retouching work is carried out to create additional product images, then product image variety is improved, but time consumption and cost increase
Solution Approach 1:
The patent performs preliminary action by training deep learning models on existing product images before needing to generate new images. The models learn the product's appearance and pose transformations in advance, enabling rapid generation of additional images without time-consuming photographing or retouching operations when actually needed
Solution Approach 2:
The system creates additional product images by copying and transforming existing images through deep learning synthesis. The neural network replicates the product's appearance in different poses by processing and regenerating image data, providing image variety without requiring separate physical photography sessions
3Ease of operation
If multiple product images are classified into categories based on pose types, then product display organization is improved, but image processing complexity increases
Solution Approach 1:
The patent replaces manual image classification operations with automated deep learning-based pose type recognition systems. The neural network models automatically analyze and categorize product images according to pose types, providing organized display structure while reducing the complexity of manual processing operations
Solution Approach 2:
The system uses deep learning models to copy and recognize pose patterns from training data to automatically classify new images. The models replicate pose type identification by comparing image features against learned patterns, enabling organized categorization without complex manual analysis
Data Source
AI summary
A method and system for creating a virtual image based on deep learning according to an embodiment of the present disclosure creates a virtual image based on deep learning by an image application executed by a processor of a computing device, where the method comprises obtaining a plurality of product images with respect to one product; classifying the obtained product images into a plurality of categories according to different pose types; determining a target category from among the plurality of categories for which the virtual image is to be created; creating a virtual image of a first pose type matched to the determined target category based on at least one product image among the plurality of product images; and displaying the created virtual image.