A Summary of Pretrained Neural Network
For engineers or scientists who work in the field of Natural language processing(NLP) or Computer Vision (CV), designing a model that works for specific task on specified data does not mean we need to start from scratch. As we know a neural network = feature extractor + task-specific layers. In computer vision, there are multiple networks that are pretrained on large image data set such as ImageNet. They are usually pretrained in an image classification task and usually be used as a head for any other specific tasks such as image semantic/instance segmentation, object detection. These network structures are commonly used as head in history: AlexNet, VGG-16/19, GoogleNet, ResNet-50/101.
Before I finish writing this blog, the following article did a good job to introduce each head.
