Resize with nearest
mode have inconsistent results compared to PyTorch and TVM
#12098
Labels
core runtime
issues related to core runtime
Describe the bug
The result is one pixel off:
when resizing a tensor
a
with shape[1, 1, 1, 1, 26]
and valuesa[...,i]=i
to[1, 1, 1, 1, 64]
with nearest mode.Urgency
None
System information
To Reproduce
The model file: model.onnx.zip
Expected behavior
They should be consistent either TVM and PyTorch are both wrong or ORT is wrong.
Screenshots
Additional context
Initiall I thought related to #12091, but now it looks like a separate issue as the inconsistent value is not constant, but just the value from the pixel right next.
The text was updated successfully, but these errors were encountered: