CODEWITHSUNDEEP
pythonnumpy
Oussama
Oussama

Reputation: 633

Change image channel ordering from channel first to last

I want to change the ordering of this numpy images array to channel_last training_data : (2387, 1, 350, 350) to (2387,350,350,1) validation_data : (298, 1, 350, 350) to (298, 350, 350, 1) testing_data : (301, 1, 350, 350) to (301, 350, 350, 1)

I tried this but it is not working

np.rollaxis(training_data,0,3).shape
np.rollaxis(validation_data,0,3).shape
np.rollaxis(testing_data,0,3).shape

Upvotes: 4

Views: 17607

Answers (3)

FlyingTeller
FlyingTeller

Reputation: 20472

You need the np.transpose method like this:

training_data = np.transpose(training_data, (0,2,3,1))

The same can be done for the other ones

Upvotes: 11

Anubhav Tripathi
Anubhav Tripathi

Reputation: 11

Do it like this:

np.rollaxis(array_here,axis_to_roll,to_where)

For your case you want to roll axis 0 to last position (beyond 3) so you have to put to_where as 4 (beyond 3).

This will work(personally tested):

np.rollaxis(train_data,0,4)

Upvotes: 0

Paul Panzer
Paul Panzer

Reputation: 53029

If the axis you are moving has length 1 a simple reshape will do:

a = np.arange(24).reshape(2, 1, 3, 4)

# Three different methods:
b1 = a.reshape(2, 3, 4, 1)
b2 = np.moveaxis(a, 1, 3)
b3 = a.transpose(0, 2, 3, 1)

# All give the same result:
np.all(b1 == b2) and np.all(b2 == b3)
# True

Upvotes: 1

Related Questions