Network in Network: Global Average Pooling
Beginner
The problem, from first principles
Instead of flattening a large feature map into a dense layer with many parameters, the paper averages each channel over space. The result is one number per class-specific channel, with no extra weights.
From theory to code
Implement global_average_pool(x), which averages an (H, W, C) feature map over its spatial dimensions and returns a length-C vector.
Constraints
- Output has shape
(C,).
Hints
Hint 1
Average over axes 0 and 1 together.
Keywords
research-papersarchitecturepooling
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.