I am currently working on an image classification task, and the process involves training with the desired target values, making it a supervised learning task.
However, when training a model, such as ResNet50, for image classification on CIFAR-10 images, can we consider the feature maps emerging after the conv5 layer as also being generated through supervised learning?
In my understanding, feature maps are optimized during supervised learning without having a clear target, suggesting that they might be considered as generated through unsupervised learning.
Therefore, should we categorize the intermediate feature maps that arise during the supervised learning process as generated through supervised learning or unsupervised learning?