Fashion Attribute Extraction via Text-to-Vector Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In the field of image processing, especially for fashion images, accurately extracting attributes from products is challenging due to the lack of information and non-standardized formats, making it difficult for consumers to evaluate products efficiently in e-commerce environments.
Innovation Solution
A method and system that utilize a computing device to train an inference model by fusing image and text information, using a text-to-vector converter to generate category and attribute vectors from text, and updating labels based on similarities, allowing the model to learn from noisy datasets without extensive manual labeling, enabling the extraction of fine-grained attributes from fashion images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual labeling is used to extract fashion attributes, then labeling accuracy can be maintained, but the time consumption and cost increase significantly
Solution Approach 1:
The system enables automatic self-labeling of fashion images through the trained inference model, which extracts attributes and categories without human intervention. The model processes images autonomously, identifying fashion items, their categories, and attributes such as color, pattern, and style, thereby eliminating the need for manual labeling while maintaining efficiency.
Solution Approach 2:
The patent replaces the mechanical manual labeling process with an automated machine learning system. The inference model, trained on labeled data, substitutes human annotators by automatically predicting fashion attributes from images, significantly reducing time consumption while preserving labeling accuracy through sophisticated image analysis algorithms.
2Reliability
If extensive manual labeling is performed to train the inference model, then model accuracy improves, but the productivity and cost decrease
Solution Approach 1:
The system performs preliminary labeling on a subset of training data to create initial labeled examples. These pre-labeled samples are used to train the inference model, which then automatically labels the remaining unlabeled data. This preliminary action approach reduces the overall manual labeling burden while still achieving high model accuracy through iterative refinement.
Solution Approach 2:
The system implements a feedback mechanism where the inference model's predictions on unlabeled data are evaluated and used to refine the model further. Labeled data is continuously added to the training set based on model performance, creating a feedback loop that improves model accuracy progressively while minimizing the need for extensive initial manual labeling.
3Loss of information
If the system processes a large number of unlabeled fashion entries, then the coverage and completeness of extracted attributes improve, but the computational complexity and resource consumption increase
Solution Approach 1:
The system segments the processing of unlabeled fashion entries into manageable batches or batches with progressive refinement. Instead of processing all entries simultaneously, the inference model processes data in stages, allowing for efficient resource utilization while still achieving comprehensive attribute extraction across the entire dataset through iterative processing.
Data Source
AI summary
A system and a method for training an inference model using a computing device. The method includes: providing a text-to-vector converter; providing the inference model and pre-training the inference model using labeled fashion entries; providing non-labeled fashion entries; separating each of the non-labeled fashion entries into a target image and target text; converting the target text into a category vector and an attribute vector using the text-to-vector converter; processing the target image using the inference model to obtain processed target image and target image label; comparing the category vector to the target image label; when the category vector matches the target image label, updating the target image label based on the category vector and the attribute vector to obtain updated label; and retraining the inference model using the processed target image and the updated label.