Count images in an ImageDataBunch

Part 1 (2019)
1

Hey I hope I’m just missing out on something here. I want to run lessons 1 dataset, but the images went through a costum filter I made. I have all of the images in a folder called ‘Data’ and their names are formatted like this : “Abyssinian_23.png”.
I load them like this:

pat = ‘/([^/]+)_\d+.png$’
data_filtered = ImageDataBunch.from_name_re(path_img, fnames, pat, ds_tfms=get_transforms(), size=224, bs=bs)

Now when I run the model and look at the confusion matrix - I sum the array for the matrix and get 295. When I do the same process with untaring the data for the original set I get 1478.
Does anyone have some idea on how to solve this?

2

The confusion matrix (and ClassificationInterp) work with your validation set only by default.

3

Thanks

So why does the validation set size is 1478 in Lesson 1 example and 295 for me? where can I control is?

4

Are you 100% certain you didn’t miss any images when moving them over, and your number of classes look correct?

5

Yes I counted the images in the folder they sum to 1478
I am suspectin the regular expression but I am not sure…

6

I am now trying to put each class in it’s own folder and try the create image databunch from folder, Is there a way to do so without the ‘train’‘test’‘validation’ folders? But still creating a validation set still?

7

You can do a split_by_rand_pct to do so. See the lesson 2 notebook. That’s how that data is set up