Viet-Anh on Software Logo

What is: RegNetY?

SourceDesigning Network Design Spaces
Year2000
Data SourceCC BY-SA - https://paperswithcode.com

RegNetY is a convolutional network design space with simple, regular models with parameters: depth dd, initial width w_0>0w\_{0} > 0, and slope w_a>0w\_{a} > 0, and generates a different block width u_ju\_{j} for each block j<dj < d. The key restriction for the RegNet types of model is that there is a linear parameterisation of block widths (the design space only contains models with this linear structure):

u_j=w_0+w_aju\_{j} = w\_{0} + w\_{a}\cdot{j}

For RegNetX we have additional restrictions: we set b=1b = 1 (the bottleneck ratio), 12d2812 \leq d \leq 28, and w_m2w\_{m} \geq 2 (the width multiplier).

For RegNetY we make one change, which is to include Squeeze-and-Excitation blocks.