In below figure, an input of (256 x 256 x 3) is given to CNN layers,...

  

Transcribed Image Text:

In below figure, an input of (256 x 256 x 3) is given to CNN layers, calculate the sizes of each features at each layer. Notice that "MP" and "FC" means "Max pooling" and "Fully connected" layers respectively. a. What is the feature size at "C1" layer, where the features at "C1" layer has been made after convolution of input of (256 x 256 x 3) with 64 filters having each size of (7 x 7), stride=5, padding = 2? b. What is the feature size at "C2" layer, where the features at "C2" layer has been made after convolution with 128 filters having each size of (5 x 5), stride=2, padding=2? c. What is the feature size at "C3" layer, where the features at "C3" layer has been made after max-pooling the features at C2 layer with a (3 x 3) max-pooling window, stride=1, padding = 0, dilation=1? d. How many parameters must be trained for the weights from C3 layer to FC layer including biases for each node in FC? Notice that FC layer has 1024 nodes in one dimensional size. 256 C2 Taboat INPUT 256 STRIDE 2000+000 1024 In below figure, an input of (256 x 256 x 3) is given to CNN layers, calculate the sizes of each features at each layer. Notice that "MP" and "FC" means "Max pooling" and "Fully connected" layers respectively. a. What is the feature size at "C1" layer, where the features at "C1" layer has been made after convolution of input of (256 x 256 x 3) with 64 filters having each size of (7 x 7), stride=5, padding = 2? b. What is the feature size at "C2" layer, where the features at "C2" layer has been made after convolution with 128 filters having each size of (5 x 5), stride=2, padding=2? c. What is the feature size at "C3" layer, where the features at "C3" layer has been made after max-pooling the features at C2 layer with a (3 x 3) max-pooling window, stride=1, padding = 0, dilation=1? d. How many parameters must be trained for the weights from C3 layer to FC layer including biases for each node in FC? Notice that FC layer has 1024 nodes in one dimensional size. 256 C2 Taboat INPUT 256 STRIDE 2000+000 1024

Answer rating: 100% (QA)

a The feature size at C1 layer is 2567451 50 b The feature size at C2 layer is 50... View the full answer