Michele LaurelliBlog

Inception Module

CNN building block applying multiple filter sizes in parallel and concatenating results.

In italiano: Inception Module

Captures features at different scales simultaneously. Uses 1x1, 3x3, 5x5 convolutions plus pooling. Reduces parameters with 1x1 bottleneck convolutions.

Examples

  • GoogLeNet/Inception v1
  • Multi-scale feature extraction
  • Efficient CNN design