How to address "UserWarning: ... for interpolate/upsample" issue with Mask R-CNN?

vision
1

Hello Forum!

I am experimenting with torchvision’s Mask R-CNN model, and it issues
a UserWarning during the forward pass:

UserWarning: The default behavior for interpolate/upsample with float
scale_factor changed in 1.6.0 to align with other frameworks/libraries,
and uses scale_factor directly, instead of relying on the computed
output size. If you wish to keep the old behavior, please set
recompute_scale_factor=True. See the documentation of nn.Upsample
for details.

First, can I safely ignore this warning?

If not, what would be the most practical way to address it?

Some context:

This is something of a known issue. See for example:

(Here, the “interpolate/upsample” issue is mentioned along side the
more central “overload of nonzero” issue.)

I see this issue running, for example, pytorch/torchvision version
1.8.0.dev20201203/0.9.0.dev20201203 (as well as on a stable
version).

Note that recompute_scale_factor isn’t mentioned in the
documentation for nn.Upsample, although it is in
nn.functional.interpolate().

Thanks for any advice.

K. Frank

2

I’ve seen the same warning and was also wondering, if the code should be adapted.

CC @fmassa
Should the old behavior be used for the interpolation or is the new one alright?