Deep Learning(CS7015): Lec 11.4 CNNs (success stories on ImageNet)

lec11mod04

[@tvscharankruthik7661]

Gold standard lectures 🏅🏅🏅

[@newbie8051]

8:50 I thought there will be some great revelation here, lol

[@mlofficial9175]

are the 96 feature maps combined by pixel-wise addition?

[@manudasmd]

5.30 if filter and input has depth 3 output should have depth 1 right?

[@signalcoder2037]

Yes, that is for one filter. But if you have N filters then depth will be N.

[@manudasmd]

​@@signalcoder2037 understood. I completed the course and earned a topper certificate also😊

[@bnglr]

with filter size changes, ZFNet's first Convolution Layer is still 55x55?

[@simrangrewal9680]

Even I didn't get how

[@divyatiwari2524]

@@simrangrewal9680 He didn't mention anything about the stride.

[@signalcoder2037]

In ZFNet, the first convolutional layer has a filter size of 7x7 with a stride of 2.

[@tvscharankruthik7661]

How did the last max-pooling layer reduce third dim from 512 to 256? Isn't that wrong because max-pooling doesn't change the value of K?

[@CHANDANSAH-ew9rk]

Timestamp?

[@signalcoder2037]

Good observation! at 16:06, I think there is a typo. Since the preceding convolutional layer has 512 filters, the depth of the input to the max pooling layer is 512. Therefore, the dimensions of the output after the last max pooling layer should be 512*2*2

[@sankhadipbera9271]

typo