Pre-trained models (ResNet, VGG, EfficientNet) in Computer Vision - Cheat Sheet & Quick Revision

Recall & Review

beginner

What is a pre-trained model in computer vision?

A pre-trained model is a neural network that has already been trained on a large dataset. It can be reused to solve similar tasks without training from scratch, saving time and resources.

intermediate

What is the main idea behind the ResNet architecture?

ResNet uses 'skip connections' or 'residual connections' that let the model learn differences from the input instead of the full transformation. This helps train very deep networks without losing information.

intermediate

How does VGG differ from ResNet in design?

VGG uses a simple and uniform design with many layers of small 3x3 filters stacked one after another. It does not use skip connections like ResNet but focuses on depth and simplicity.

advanced

What is special about EfficientNet compared to older models like VGG and ResNet?

EfficientNet scales the model's depth, width, and resolution together in a balanced way. This makes it more efficient, achieving better accuracy with fewer parameters and less computation.

beginner

Why do we use pre-trained models like ResNet, VGG, and EfficientNet in new computer vision tasks?

We use them to save time and improve performance. Since they learned useful features from large datasets, they help new models learn faster and better even with less data.

What is the key feature of ResNet that helps train very deep networks?

A Very large convolution filters

B Reducing the number of layers

C Using only fully connected layers

D Skip connections that add input to output

ResNet uses skip connections to let the network learn residuals, which helps avoid problems like vanishing gradients in deep networks.

Which model uses many small 3x3 convolution filters stacked deeply?

A VGG

B ResNet

C EfficientNet

D AlexNet

VGG is known for its simple design using many small 3x3 filters stacked in deep layers.

What does EfficientNet optimize to improve performance?

A Only depth of the network

B Only width of the network

C Depth, width, and resolution together

D Only resolution of input images

EfficientNet balances scaling depth, width, and input resolution to improve accuracy and efficiency.

Why are pre-trained models useful for new tasks?

A They require more data to train

B They start with learned features from large datasets

C They always have fewer layers

D They do not need any training

Pre-trained models have learned useful features from big datasets, so they help new tasks learn faster and better.

Which of these is NOT a characteristic of VGG?

A Use of skip connections

B Deep network with many layers

C Stacked 3x3 convolution layers

D Simple and uniform architecture

VGG does not use skip connections; that is a feature of ResNet.

Explain how ResNet's skip connections help in training deep neural networks.

Think about how adding input directly to output helps the network learn.

Describe the main differences between VGG and EfficientNet architectures.

Focus on design style and scaling approach.