图像分类元模型的训练方法、图像分类方法、装置和介质
By constructing pseudo-tasks from inverted image data and training an image classification meta-model using pseudo-support sets, the problem of model training in data-free scenarios is solved, thereby improving model performance and expanding application scenarios.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JD DIGITS HAIYI INFORMATION TECHNOLOGY CO LTD
- Filing Date
- 2023-07-20
- Publication Date
- 2026-07-17
AI Technical Summary
In scenarios without data, existing technologies struggle to effectively utilize pre-trained models for training image classification models, resulting in poor model performance and making them unsuitable for pre-trained models of different structures and scales.
By inverting image data based on multiple pre-trained models, a pseudo-task is constructed, and the image classification meta-model is trained using pseudo-support sets and pseudo-query sets. The learning difficulty is dynamically adjusted by combining real-time feedback to optimize the model parameters.
It improves the performance of image classification meta-models, enabling them to be applied to pre-trained models of different structures and scales, expanding application scenarios, and improving the efficiency and accuracy of model training.
Smart Images
Figure CN116704295B_ABST