Network in Network: A Valid 2D Convolution
Intermediate
The problem, from first principles
The paper's micronetworks sit on top of ordinary spatial convolutions. This question implements that spatial part directly, so the 1x1 layers from the previous question have something to act on.
From theory to code
Implement conv2d_valid(img, kernel), which slides the kernel over the image with no padding and returns the sums of elementwise products.
Constraints
- Output size is
(H - kh + 1, W - kw + 1). - This is cross-correlation: the kernel is not flipped.
Hints
Hint 1
Use two loops over output positions and sum the elementwise product of each window with the kernel.
Keywords
research-papersarchitectureconvolution
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.