Source: Deep Learning on Medium
Pre-trained models are models that have been trained using available public dataset and was done on the basis of some research paper/finding.
Keras is most popular library in Deep Learning which contain lots of model. Backend service like Tensorflow help provide an abstraction in the use of Keras trained model. A trained model has two parts — Model Architecture and Model Weights. Some of the available trained model for image classification include (based on keras website):
- ResNet, ResNetV2, ResNeXt
The essence of this article is to decide on which model is suitable for the task at hand which is classifying of image into ‘hotel’ and ‘not-hotel’. To solve this task, we will be using transfer learning which will save us time. And transfer learning is usually expressed through the use of pre-trained models.
My suggestion is for my team to pick among “Xception or InceptionV3 model” since it has improved performance based on the statistics shown above.