一种基于对抗偏移训练的自适应图像掩码方法及系统
By employing an adaptive image masking method trained against offsets, images are processed in blocks and an adaptive masking scheme is designed. This addresses the issue of the ViT model's poor performance on small datasets, thereby improving the model's performance and robustness.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- TONGJI UNIV
- Filing Date
- 2023-09-12
- Publication Date
- 2026-07-17
AI Technical Summary
The ViT model performs poorly on small datasets, mainly due to a lack of sufficient inductive bias, making it difficult to effectively combine local information and global features.
An adaptive image masking method based on adversarial offset training is used to process the original image in blocks and perform feature transformation. An adaptive masking scheme is designed, and the model training is optimized using a weight loss function. By combining adversarial offset loss and cross-entropy loss, irrelevant information is masked, and the model's ability to extract useful information is enhanced.
It improves the performance of the ViT model on small datasets, enhances its ability to distinguish useful information, reduces interference from irrelevant information, and improves the model's robustness and generalization ability.
Smart Images
Figure CN117409236B_ABST