VGG: Receptive Field of Stacked 3x3 Convolutions
Beginner
The problem, from first principles
VGG (Simonyan & Zisserman, 2014) replaced large filters with stacks of 3x3 convolutions. Three 3x3 layers see the same area as one 7x7 layer, with fewer parameters and more non-linearities.
From theory to code
Implement receptive_field_stack(n, k), the receptive field of a stack of n convolutions with kernel k.
Constraints
- Stride one, as in the paper's stacks.
Hints
Hint 1
Each extra layer adds k - 1 pixels to the field, starting from one pixel.
Keywords
research-paperscomputer-visioncnnarchitecture
Python
Run your code to see results
No results yet
Run your code to check it against the examples, or submit to run every test.