How to dynamically augment data

[RuoyuChen10]

I hope to dynamically enhance the data, such as calculating the saliency map through Grad-CAM, and then take some measures to enhance it in combination with the current input. How should this method be implemented in programming? It seems that after updating the model parameters, use the current parameters to calculate the saliency map, then combine it with the input, and then use these data to update the model, and so on.

It would be great if you can find the function of the for loop. Similar to the general classification model, you first get the data from the dataloader, then input the model to calculate the loss, update and then calculate the enhanced data, and then use the enhanced data to update the current detector. I don’t know where to write it. Obviously not suitable for writing in dataloader.