I am currently studying deep metric learning. I use Cars-196 dataset and save the model in each iteration or epoch to check how data and proxies are updated. But there are quite many classes so I think that it’d be better to reduce classes, save and visualize them.
What is better way to modify for saving specific classes?
- reduce the model before training: it means, for example, train only 10 classes
- save the model after training but only save 10 specific classes: train all the 98 classes and just save 10 specific classes
- save all the data but visualize only 10 specific classes.