Skip to main content

> ML_LITERATURE // SZEGEDY-2015-GOING-DEEPER-WITH-CONVOLUTIONS-GOOGLENET-INCEPTION_v1.0

Going Deeper with Convolutions (GoogLeNet / Inception)

Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed, Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke, Andrew Rabinovich · IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2015)

seminal-architecture2015foundationalthirdPartyReproduced
DOI: 10.1109/CVPR.2015.7298594Back to Literature

Principal Contribution

Introduced the Inception module with multi-scale parallel convolutional filters (1x1, 3x3, 5x5) and 1x1 dimension-reduction bottlenecks.

Operational Relevance

Inception-v3 backbone remains the universal reference feature extractor for computing Fréchet Inception Distance (FID) in generative AI.

Assumptions

  • Optimal sparse network structures can be approximated by dense, readily available computational building blocks

Limitations

  • Complex non-uniform branching structure difficult to optimize for high-throughput GPU memory bandwidth compared to linear ResNets

Connected Algorithms, Architectures & Tools

Related Algorithms:
Simple Framework for Contrastive Learning (SimCLR)
Related Architectures:
Implementing Libraries: