References
- Udacity (2016. 6. 6.). Inception Module. YouTube. [LINK]
- Udacity (2016. 6. 6.). 1×1 Convolutions. YouTube. [LINK]
- Tommy Mulc (2016. 9.?25.). Inception modules: explained and implemented. [LINK]
- Szegedy et al. (2015). Going Deeper with Convolutions. CVPR 2015. [arXiv]
Summary
History
- The inception module was first introduced in?GoogLeNet for?ILSVRC’14 competition.
Key concept
- Let?a convolutional network decide effective computation among 1×1 convolutions, 3×3 convolutions, 5×5 convolutions, and 3×3 max pooling.
The architecture of an inception module
- Yellow boxes are used for depth reduction in order to reduce computation complexity.