VGG: Multiply-accumulates of a Convolution
Intermediate
The problem, from first principles
VGG's cost is dominated by its convolutions. Counting multiply-accumulates per layer tells you where the compute goes, which matters when comparing it with later architectures.
From theory to code
Implement conv_macs(h, w, c_in, c_out, k), the multiply-accumulate count of one convolution.
Constraints
- Default kernel is 3.
Hints
Hint 1
Multiply the output area, input channels, output channels and kernel area.
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.