Why not just do the train_test_split over directory names, and later (e.g. in a Pipeline) read in the images? On 15 January 2018 at 10:18, Abdul Abdul <abdul.sw84@gmail.com> wrote:

Hello,

I'm trying to train an image classifier, but a bit confused on how to label my data. The issue here is that for each class I have subdirectories, each of which contains two images. So, it is not I have classes, and in each class I simply have the images that come under that class (i.e. cats vs. dogs).

I will show here some attempts for grouping the data together, but not yet able to figure how to assign the label, and pass the pairs of images along with the label to the image classifier.

So, that's how I simply read the two images:

im1 = cv2.imread('img1.jpg') im1 = img_to_array(im1)

im2 = cv2.imread('img2.jpg') im2 = img_to_array(im2)

I then *pair* the images as follows:

pair = (im1,im2)

For labeling, this is what I did:

label = root.split(os.path.sep)[-2] label = 1 if label == 'cat' else 0

How can I group the above pairs of images (im1,im2) and attach the label to them? Especially that I want to pass them to the following scikit-learn function:

(trainX, testX, trainY, testY) = train_test_split(data, labels, test_size=0.25, random_state=42)

Thanks.

_______________________________________________ scikit-learn mailing list scikit-learn@python.org https://mail.python.org/mailman/listinfo/scikit-learn