Multiactivation Pooling Method in Convolutional Neural Networks for Image Recognition

Table 5

The baseline ALL-CNN architecture ‎[24] and our modified conv-maxpool/MAP architectures.

conv-maxpool/MAP		All-CNN
Input (3232 RGB image)
conv3-96	conv3-96	conv3-96
conv3-96	conv3-96	conv3-96
conv3-96	conv3-96
maxpool/MAP 44	conv3-192	conv3-96 stride=2
conv3-192	conv3-192	conv3-192
conv3-192		conv3-192
maxpool/MAP 44	maxpool/MAP 88	conv3-192 stride=2
conv3-192	conv3-192	conv3-192
conv3-192	conv3-192	conv1-192
		conv1-10
avgpool 22	avgpool 44	avgpool 66
fc1 192192	fc1 192192	Softmax-10
fc2 19210	fc2 19210
Softmax-10	Softmax-10