yep, that's good if you want to retrain a pre-trained model for specific categories on new image data (that are pretty close to imagenet type images)