RuntimeError: invalid argument 0: Sizes of tensors must match except in dimension 1.

[NTNguyen13]

Hi, I'm currently using this package for segmentation of multichannel confocal microscopy cell image. The package works well for most of the case, except when image size is (1728, 1728). I have tried it with different size such as (2048, 2048) and (3072, 3072), it's still good.

To reproduce the problem:

NUC_MODEL = '~/HPA_image/segmentation_weight/dpn_unet_nuclei_v1.pth'
CELL_MODEL = ~/HPA_image/segmentation_weight/dpn_unet_cell_3ch_v1.pth'

segmentator = cellsegmentator.CellSegmentator(NUC_MODEL,
                                              CELL_MODEL,
                                              scale_factor=0.25,
                                              device='cpu',
                                              padding=False,
                                              multi_channel_model=True
                                              )

nuc_segmentations = segmentator.pred_nuclei(path_nu)
cell_segmentations = segmentator.pred_cells(path_3channel)

I have tried with multiple images, it always return the same error: RuntimeError: invalid argument 0: Sizes of tensors must match except in dimension 1. Got 28 and 27 in dimension 2 at /pytorch/aten/src/TH/generic/THTensor.cpp:612

if the image size is (1728, 1728).

Could you please help me with this? Thank you

[Linsho]

I think that problem can be solved by setting the value of padding to True.

[cwinsnes]

Yes, if you set padding to True this should be fixed. Could you verify that's the case?

[NTNguyen13]

yes, it's fixed using padding=True