Deep Learning - Convolutional Neural Networks with TensorFlow - Some Pre-Trained Models (VGG, ResNet, Inception, MobileN

The video tutorial covers various pre-trained models used in transfer learning, including VGG, Resnet, Inception, and Mobile Net. It explains the architecture and variations of each model, highlighting their unique features and use cases. The tutorial also emphasizes the importance of data preprocessing and provides guidance on using Tensorflow and Keras functions to handle input data formats.

What is the primary characteristic of the VGG network that differentiates it from earlier CNNs?

• It uses a single convolutional layer.

• It has a large number of layers.

• It operates on grayscale images only.

• It is designed for mobile devices.

What is a unique feature of the Resnet architecture?

• It has no fully connected layers.

• It includes branches for learning residuals.

• It only uses 3x3 filters.

• It uses a single branch for learning.

How does the Inception network handle different filter sizes?

• It applies multiple filter sizes in parallel.

• It uses only the largest filter size.

• It alternates filter sizes in each layer.

• It uses a single filter size for all layers.

What is the main advantage of using MobileNet?

• It uses the largest number of layers.

• It requires no preprocessing of input data.

• It is lightweight and suitable for mobile devices.

• It is highly accurate for all tasks.

Why is it important to match the input data format when using pre-trained CNNs?

• To avoid errors during model training.

• To maintain the model's accuracy.

• To reduce the model's computational cost.

• To ensure compatibility with the model's architecture.

What preprocessing step is specific to VGG's input data?

• Scaling pixel values to -1 to 1 range.

• Normalizing pixel values to 0-1 range.

• Converting images to grayscale.

• Using BGR color order.

Which function in TensorFlow and Keras helps with input preprocessing for models like VGG and Resnet?

• preprocess_data

• data_scaler

• input_normalizer

• preprocess_input