5. A user has collected a lot of 100 x 100 size images. Some of them...

Transcribed Image Text:

5. A user has collected a lot of 100 x 100 size images. Some of them contain mountains and the rest are of grass field. As the number of images was large and he did not want to manually label all the images. He has devised a neural network as shown below to do the same. Image (100x100x1) CNN F = 32 Max-pool (2x2) CNN F = 32 Max-pool (2x2) Max-pool (2x2) CNN F = 6A Max-pool (2x2) CNN F = 64 Fully Connected (32) Fully Connected (2) All the activation functions are RelU, except at the final layer, where it is softmax. Kernel size for all the CNNs is kept throughout 5 x 5. Determine the number of trainable parameters for this architecture? Assume bias term need not to be added for the CNN kernels and the image size reduces while applying a convolution operation. Also determine size of the output at every convoution step. 5. A user has collected a lot of 100 x 100 size images. Some of them contain mountains and the rest are of grass field. As the number of images was large and he did not want to manually label all the images. He has devised a neural network as shown below to do the same. Image (100x100x1) CNN F = 32 Max-pool (2x2) CNN F = 32 Max-pool (2x2) Max-pool (2x2) CNN F = 6A Max-pool (2x2) CNN F = 64 Fully Connected (32) Fully Connected (2) All the activation functions are RelU, except at the final layer, where it is softmax. Kernel size for all the CNNs is kept throughout 5 x 5. Determine the number of trainable parameters for this architecture? Assume bias term need not to be added for the CNN kernels and the image size reduces while applying a convolution operation. Also determine size of the output at every convoution step.