一种图像分类模型的训练方法及装置
By freezing the backbone network parameters of the image classification model and adjusting only the feature cue word vectors and classifier parameters, the problem of high computational overhead in training large image classification models is solved, achieving a training method that improves model accuracy and saves resources.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING LONGZHI DIGITAL TECH CO LTD
- Filing Date
- 2023-02-08
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies suffer from high computational overhead and limited improvement in model accuracy during the training of large-scale image classification models, especially when fine-tuning the pre-trained backbone network, where resource consumption is high and the results are unsatisfactory.
An image classification model training method is adopted, which freezes the parameters of the backbone network and adjusts only the parameters of the first feature prompt word vector and the classifier. By adding learnable feature prompt word vectors to adapt to the distribution of training sample data, the model accuracy is improved.
It reduces the computational overhead of model training, improves model accuracy, and has low computational complexity and data computation volume, resulting in relatively low cost and achieving fine-tuning training with low resource consumption.
Smart Images
Figure CN116188854B_ABST