Convolution Neural Network, Image Category Classification Using Deep Learning example

Hello I have recently started working Convolution Neural Network. I found a tutorial https://au.mathworks.com/help/nnet/convolutional-neural-networks.html I have confusion in the "Image Category Classification Using Deep Learning" example. The example uses pre-trained model which is developed for the 1000 class. is it possible to use the model of 1000 class for 3 class?

[Cy Pi]

Yes, it's possible to use the model of 1000 class for 3 class. You just remove the last layer that has the 1000 classes and retrain it with 3 nodes at the last layer. There is a tool called Hive Moderation https://hivemoderation.com/ that does this.

Answer 1

[Walter Roberson]

No. The pretrained model already has the information about the other 997 classes built in, and you cannot just extract a subset of the information for 3 classes. If you wanted that you should extract the three classes from the original data and train with it.

You can use the pretrained model and apply it only to images intended to be in the three classes, but you cannot be sure it will not decide that the image belongs to one of the other classes.

[Adnan Farooq Awan]

I have a different data set and the total number of classes are 10. In that case, do I need to make my own architecture or how can I use this one?

[Walter Roberson]

You need to train on your data, but the architecture might end up looking the same