When attempting to predict a single image I get a "operands could not be broadcast together" error

Part 1 (2018)
1

In working through the lesson 1 stuff I wanted to try using my trained dataset on a separate single test image.

I found several topics on how to do that here, but am getting an error I didnt see mentioned. Here is the code im using:

trn_tfms, val_tfms = tfms_from_model(arch,sz) # get transformations
im = (np.array(Image.open(f’{PATH}testimage1.png’)))
print(im.shape)
im = val_tfms(im)
learn.precompute=False # We’ll pass in a raw image, not activations
preds = learn.predict_array(im[None])
np.argmax(preds) # preds are log probabilities of classes

Note that I broke that up a bit in an effort to understand the error. The same error occurs when condensed into the versions posted on this forum.

Here is the error that occurs on the val_tfms(im) step:

Capture
Capture.PNG1154×715 43.2 KB

2

It looks like your image has 4 channels. Is that expected?

3

Its the same kind of images I am using in my training data. Just one of them I pulled from that actually. Its possible it has 4 by design.

Is a 4th channel an issue for some reason?

4

turns out that was the case. Not sure why it caused problems but trained fine on the same images. converting everything to jpg seems to have fixed it

5

Try to use open_image(f’{PATH}testimage1.png’)) instead of Image.open(f’{PATH}testimage1.png’) - this is what fastai uses under the hood.