For most computer vision problems, you can usually better off using a pre-trained model such as AlexNet, VGGNet, GoogleNet, ResNet etc. Let’s study how exactly one should go about doing this.
Thus, the initial layers of a network extract the basic features, the latter layers extract more abstract features, while the last few layers are simply discriminating between images.
In other words, the initial few layers are able to extract generic representations of an image and thus can be used for any general image-based task. Let’s see some examples of tasks we can use transfer learning for.
Thus, transfer learning is not just limited to image classification but can be extended to a wide variety of tasks. In the next segment, you will learn how to train pre-trained models for specific purposes.
Report an error