Beginning with MATLAB R2021a, the documentation no longer lists these options as unsupported in the trainFasterRCNNObjectDetector function. You can refer to the R2021a documentation here and the latest documentation here.
To prevent overfitting of a network without using a validation set, consider the following strategies:
- Manual Validation Split: Before training, manually split your dataset into training and validation sets.
- Cross-Validation: Use k-fold cross-validation to train and evaluate your model multiple times on different subsets of data.
- Regularization Techniques: Apply techniques such as dropout, data augmentation, and weight decay to prevent overfitting.
- Ensemble Methods: Train multiple models and average their predictions. This can sometimes reduce overfitting by balancing out the individual model biases.
Refer to these resources for further reading on Overfitting:
