Research Article
An Improved Self-Training Model with Fine-Tuning Teacher/Student Exchange Strategy to Detect Computer-Generated Images
Algorithm 1
Improved self-training algorithm with FTTSE.
| Input: labeled dataset & unlabeled dataset | | Output: model | (1) | Train teacher model with labeled dataset . | | | (2) | Use the trained teacher model to predict the pseudo label on unlabeled dataset . | | | | And then combine the pseudo-labeled dataset and the labeled dataset together to generate a mixed dataset. | (3) | Apply malicious attacks to to get attacked mixed image dataset | (4) | Train the student model with the attacked mixed image dataset . | | | (5) | | (6) | while neural network unfitting do | (7) | Last student model is regarded as the teacher model to predict pseudo label and generate a new attacked mixed dataset | | | (8) | Last teacher model trained from previous round of iteration: is fine-tuned and used for retraining student model with dataset . | | | (9) | Implement the learning rate decay strategy to reduce the learning rate value as follows: | | | (10) | | (11) | end while |
|