Hello,
I am training a unet model, which is fine. am also able to load the checkpoint and resume training .
Say i train the model on Animal datasets up to 100 epochs. Then i want to train Car Datasets after , do i need to resume training with both datasets mixed ? or i can only add car datasets and the next checkpoint to be saved will have learned on both ?
Thank You
Generally speaking there is no fixed recipe you have to follow here, you can train on both datasets in any order or with different degrees of mixing. It’s hard to know ahead of time what will give the best results as it is likely dataset and use-case dependent. The only thing to pay attention to is how the last layers are swapped in an out (as I imagine the classes in Cars and Animals are different and there are different numbers of classes between the datasets).