AI Post Transformers • Visualization Companion

Caffe and the Rise of CNN Frameworks

A systems-focused visual companion to the episode: how Caffe made early CNN research runnable, repeatable, and portable enough to become infrastructure instead of one-off lab code.

Primary Paper

Jia et al., 2014
arXiv:1408.5093

Core Claim

Separate model description from implementation so CNN experiments and deployment stop being different codebases.

Key Abstractions

Protocol Buffers configs, DAG network structure, and blobs that hide CPU versus GPU storage.

Historical Signal

Quoted throughput above 40M images/day on one K40 or Titan GPU, plus pretrained model distribution.

Primary arXiv ID • 1408.5093 Transcript arXiv IDs extracted • 1408.5093 Interactive Viz Link

References

Caffe: Convolutional Architecture for Fast Feature Embedding Yangqing Jia, Evan Shelhamer, Jeff Donahue, Sergey Karayev, Jonathan Long, Ross Girshick, Sergio Guadarrama, Trevor Darrell, 2014.
https://arxiv.org/abs/1408.5093

ImageNet Classification with Deep Convolutional Neural Networks Alex Krizhevsky, Ilya Sutskever, Geoffrey E. Hinton, 2012.
Google Scholar

OverFeat: Integrated Recognition, Localization and Detection using Convolutional Networks Pierre Sermanet et al., 2013.
Google Scholar

DeCAF: A Deep Convolutional Activation Feature for Generic Visual Recognition Jeff Donahue et al., 2014.
Google Scholar

cuda-convnet Alex Krizhevsky, 2012.
Google Scholar

Theano / Pylearn2 / Torch7 Contemporary frameworks that formed the 2011–2014 ecosystem Caffe entered.
Theano • Pylearn2 • Torch7

Caffe and the Rise of CNN Frameworks

References

Listen to the Episode