Definition Recognition Training With Unsupervised Pre-Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The high time and labor costs of manual annotation in training definition recognition models result in limited data sets and low accuracy of content definition recognition.
Innovation Solution
A method involving unsupervised pre-training on a feature extraction network using a pre-training content set, followed by definition recognition with a prediction network, and updating model parameters based on training loss to improve recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual annotation is used to train the definition recognition model, then the model can be trained with labeled data, but the time cost and labor cost are very high, limiting the data set size
Solution Approach 1:
The patent applies unsupervised pre-training before supervised training. The feature extraction network is first pre-trained on a large corpus of unlabeled images to learn general visual features, then fine-tuned on a smaller set of manually annotated definition recognition data. This preliminary unsupervised training enables the model to leverage large amounts of unlabeled data, significantly reducing the time and labor costs associated with manual annotation while maintaining high definition recognition accuracy.
2Measurement precision
If manual annotation is used to train the definition recognition model, then the model can be trained with labeled data, but the labor cost is very high, limiting the data set size
Solution Approach 1:
The patent applies unsupervised pre-training before supervised training. The feature extraction network is first pre-trained on a large corpus of unlabeled images to learn general visual features, then fine-tuned on a smaller set of manually annotated definition recognition data. This preliminary unsupervised training enables the model to leverage large amounts of unlabeled data, significantly reducing the time and labor costs associated with manual annotation while maintaining high definition recognition accuracy.
Solution Approach 2:
The patent introduces an intermediary unsupervised pre-training stage between data collection and supervised training. This intermediary stage processes large amounts of unlabeled data to pre-training the feature extraction network, acting as a mediator that bridges the gap between abundant unlabeled data and limited labeled data, thereby enabling effective training with smaller annotated data sets.
3Measurement precision
If a larger data set is used for training, then the definition recognition accuracy can be improved, but the time cost and labor cost of manual annotation increase
Solution Approach 1:
The patent applies unsupervised pre-training before supervised training. The feature extraction network is first pre-trained on a large corpus of unlabeled images to learn general visual features, then fine-tuned on a smaller set of manually annotated definition recognition data. This preliminary unsupervised training enables the model to leverage large amounts of unlabeled data, significantly reducing the time and labor costs associated with manual annotation while maintaining high definition recognition accuracy.
Data Source
AI summary
A training method includes obtaining a pre-training content set, sample content, and a definition level annotation corresponding to the sample content, performing unsupervised pre-training on a feature extraction network of a definition recognition model based on content in the pre-training content set to obtain a pre-trained feature extraction network, performing definition recognition on the sample content using the pre-trained feature extraction network and a prediction network of the definition recognition model to obtain a predicted definition level, calculating a training loss based on the predicted definition level and the definition level annotation, and updating one or more model parameters of at least one of the pre-trained feature extraction network and the prediction network according to the training loss.