Hello!
I am having this issue in the training when extracting the prediction:
output = model.forward(image)
The shape of the image tensor is: torch.Size([1, 1, 299, 299])
Batch size: 1
Image size: 299, 299
Grayscale
The model is the inception v3
Anyone has any idea why this is happening?
Thank you in advance
Hi. When I try to reproduce the behavior you’re describing:
m = torchvision.models.inception_v3(init_weights=False) # this is for the speedup
img = torch.randn(1,1,299,299) # dummy image
- If use image with
torch.Size([1, 1, 299, 299])
, I got this error:
RuntimeError: Given groups=1, weight of size [32, 3, 3, 3], expected input[1, 1, 299, 299] to have 3 channels, but got 1 channels instead
The model doesn’t work with 1 channel images
- If you want to predict on the batch size=1, you should put the model in evaluation mode.
- To make it work with grayscale image you can try to just copy 1 channel into 2 additional, so to have the correct number of channels and see if it provides meaningful results to you.
This post on how to deal with grayscale image may be of help to you.
I actually had forgotten to uncomment the line that changed the first layer of the inception to receive grayscale images ! Thank you for the help!
1 Like