AI Model Pre-training with Self-Supervised Image Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The fashion industry faces challenges in continuously updating AI models to respond to changing trends due to the time-consuming and resource-intensive process of preparing large amounts of high-quality training data for supervised learning, and existing self-supervised training methods are limited to specific fields like road area estimation.
Innovation Solution
A method and system for automatically improving AI model performance through self-supervised pre-training using image data, including fashion items, and fine-tuning with image and text information, allowing for continuous data addition and adaptation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If supervised training is used to train AI models with labeled answer data, then the training process becomes easier and more stable with clearer performance evaluation, but a large amount of high-quality training data is required which takes significant time and human resources to prepare
Solution Approach 1:
The patent applies pre-training on unlabeled data before fine-tuning with labeled data. This preliminary action of training on abundant unlabeled fashion images prepares the model in advance, reducing the time and resources needed for subsequent supervised training with limited labeled data.
2Ease of operation
If supervised training is used to train AI models with labeled answer data, then the training process becomes easier and more stable with clearer performance evaluation, but a large amount of high-quality training data is required which consumes significant human resources
Solution Approach 1:
The patent employs self-supervised learning where the model learns from unlabeled data without requiring extensive human annotation. The system automatically extracts features and patterns from unlabeled fashion images, reducing the need for human resources in data labeling and preparation while maintaining training effectiveness.
3Adaptability or versatility
If existing self-supervised training methods are used, then training on unlabeled data is enabled, but the methods are limited to specific fields like road area estimation and cannot be applied to fashion industry AI models
Solution Approach 1:
The patent develops a universal pre-training framework that can be applied across different AI model types in the fashion industry, including item recognition, attribute recognition, and location detection models. The method extracts general-purpose features from unlabeled fashion images that are transferable to multiple specific tasks, overcoming field-specific limitations of existing self-supervised methods.
Data Source
AI summary
A method of automatically improving artificial intelligence (AI) model performance is provided. The method includes: collecting image data including at least one item; performing pre-training based on self-supervised training for a pre-training model by using the image data; and setting initial values of the AI model and performing fine-tuning by using the pre-training model.