Pretrained models are used because they offer a shortcut to building accurate models. They are already trained on massive datasets, learning rich patterns and representations. By using pretrained models, we can save time and effort required for training from scratch. These models act as a solid foundation, providing good initial weights and feature extractors. They are particularly useful when we have limited data, as they transfer their knowledge to new tasks. With pretrained models, we can achieve impressive results with less training time and data, making them a valuable resource in the field of machine learning.

Pretrained models in keras:

In Keras, you can utilize various pretrained models for CNN tasks through the keras.applications module. This module provides popular pretrained models that have been trained on large-scale datasets. Here are a few examples:

1. VGG16: from keras.applications import VGG16 This model is based on the VGGNet architecture and is known for its depth and performance on image classification tasks.

2. ResNet50: from keras.applications import ResNet50 ResNet50 is a deep residual network that achieved breakthrough results in the ImageNet competition. It is effective for both classification and feature extraction tasks.

3. InceptionV3: from keras.applications import InceptionV3 InceptionV3 is a powerful model that utilizes inception modules for efficient feature extraction. It performs well on various visual tasks.

4. MobileNet: from keras.applications import MobileNet MobileNet is a lightweight model designed for mobile and embedded devices, offering a good trade-off between performance and model size.