From: The effect of whitening transformation on pooling operations in convolutional autoencoders
Pooling size | Pooling type | Unwhitened | Whitened ( ε = 0.1) |
---|---|---|---|
8 × 8 | Avg | 59.84% | 64.81% |
Max | 60.55% | 63.62% | |
Ada | 60.67% | 65.48% | |
Sto | 61.06% | 64.44% | |
5 × 5 | Avg | 57.34% | 64.38% |
Max | 59.56% | 63.06% | |
Ada | 58.66% | 63.59% | |
Sto | 58.60% | 63.25% |