liveBook · Manning

ImageNet

As explained in Inside Deep Learning: Math, Algorithms, Models

A large image classification dataset with 1,000 classes used here as the pretraining source domain for ResNet.

ImageNet is a large image-classification dataset with 1,000 classes and about 1 million training images. It serves as the source domain for pretraining ResNet before adapting the model to a new classification task. section 13.2.1

Pretraining gives ResNet's convolutional filters a useful starting point: they already detect broadly reusable features such as edges, colors, sharpening, and blurring. Replacing the final fully connected layer leaves that layer randomly initialized for the target labels. section 13.2.1

How Inside Deep Learning: Math, Algorithms, Models covers it

How deep it goes
Explains dataset scale, pretrained filters, and classifier replacement, but does not detail ImageNet's data construction.
Examples
Uses a pretrained ResNet-18 adapted with a new two-class layer for cat-versus-dog classification.
What it assumes
Assumes a ResNet classification workflow and a separate target task, while treating convolutional feature reuse as the rationale.
What you can do after
Apply via torchvision.models by loading an ImageNet-pretrained ResNet and replacing its final classifier.
How it connects
Connects ImageNet to transfer learning, ResNet, convolutional neural networks, source domains, and target domains.
Cover of Inside Deep Learning: Math, Algorithms, Models
liveBookmanning.com
Read Inside Deep Learning: Math, Algorithms, Models on liveBook Get the book at manning.com

Also explained in

Compare how 2 books explain ImageNet →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Inside Deep Learning: Math, Algorithms, Models. The section links point to the parts of the book it draws on.