C4 W2 MobileNet Architecture

Course Q&A Deep Learning Specialization Convolutional Neural Networks
1

image
image1827×862 90.9 KB

Is it right to say that in this MobileNet v2 Bottleneck, the Expansion and Pointwise operations are 1x1 Convolution Operations?

Also, the same can be said for Depthwise Separable Convolution. Where the first step is Depthwise Convolution and the second step is 1x1 Convolution a.k.a Pointwise Convolution.

Is this right?

2

Hey @Ammar_Jawed,

Yes you are correct!.

In MobileNetV2:

  • Expansion and Pointwise: It uses a 1x1 convolution to increase or decrease the dimensionality.
  • Depthwise Separable Convolution: It involves Depthwise Convolution (e.g., 3x3 or 5x5) followed by Pointwise Convolution (1x1 Convolution) to capture and mix features.

Regards,
Jamal

3

What is the name of this course and is it on coursera…?

4

@Chirayu_Sharma Yes, this is from course 4 of 5 of the Deep Learning Specialization, found here:

Convolutional Neural Networks Course by DeepLearning | Coursera

However, unless you are already really familiar with NN structure/calculations and already know a bit of Tensorflow, I’d highly suggest you take the 3 preceding courses and then the fourth in that order.

2 Likes