Week 2: Downloading open source code vs using TensorFlow

Hi, in week 2, Professor Ng mentioned that we can use open-source implementation when implementing ConvNet. I am fairly new to using open-source code. I was wondering what are the differences between using open-source implementation and calling a python library such as TensorFlow? Thanks!

Hello @Marcia_Ma,

Open-source means the source code was made freely available. Tensorflow is an open-source library and you can find its source code here.

If you have built a computer vision neural network using Tensorflow, and decided to publish its source code to make it available to everyone too, then your neural network is open-sourced too.

If I have to build a computer vision neural network like yours, then I have at least two choices:

  1. I use the open-source Tensorflow framework to build it from the ground up
  2. I use your open-source neural network as the starting point and build it from there

The difference between 1 and 2 is that approach 2 takes me less time on development because it is based on your existing work.

Cheers,
Raymond

PS: If you also provided pre-trained weights, and I start my training from your pre-trained weights (instead of randomly initialized weights), this style of training is called “Transfer Learning”. The topic of transfer learning was covered in Course 3 Week 2.

1 Like

Thank you Raymond! That was really helpful!