Using a target size that is different to the input size is deprecated. Please ensure they have the same size

vision
1

Hi, Doing Background Removal With U2net.
The error i’m getting is :

Using a target size (torch.Size([5, 3, 320, 320])) that is different to the input size (torch.Size([5, 1, 320, 320])) is deprecated. Please ensure they have the same size.

As you may know, This is a Segmentation Type task. so, it has RGB Image as X and Ground Truth Grayscale Image as Y.

The error is when i apply this model output to Loss Function. I’m Using BCE Loss.

What should i do? i don’t want to debug that error by just converting input image to Grayscale.
is there any other way ?

Thanks in advance.

2

your input size is (5,1,320,320) the model is expecting to have (5,3,320,320).

If you don’t want to pass RGB images

you can change
Conv2d(3,-,(-,-)) to Conv2d(1,-,(-,-))

if you are using a linear layer you can change

Linear(3*320*320,-) to Linear(1*320*320)

Hope this helps

With best regards,
Ranuga

1 Like
3

And also I have seen this more in BCELoss (In my personal Projects)

I am not sure if it has anything to do with BCELoss.

4

Hi,

Thanks for your reply.

What you have said make sense in spite of the model but the problem was in processing the image Data. I could have explained that extensively.

Again, Thanks for the reply.

1 Like