You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently we have the adapt_batches configuration setting which allows the _adapt_batches function to modify the batch shape, I have recently through a configuration mistake noticed if you supply a sample that includes NWPs with width and height smaller than the output size that it will pad the NWP data in the samples with 0 values, as per torchvisions center_crop documentation .
This is done silently currently, I think we should either explicitly throw an exception for this case of input size smaller than requested output size as I don't see it as a case we want to support/part of the original intention of this function or at least throw a warning so it's more obvious this is happening, keen to hear thoughts on this, thanks! @peterdudfield@dfulu@AUdaltsova
The text was updated successfully, but these errors were encountered:
Currently we have the
adapt_batches
configuration setting which allows the _adapt_batches function to modify the batch shape, I have recently through a configuration mistake noticed if you supply a sample that includes NWPs with width and height smaller than the output size that it will pad the NWP data in the samples with 0 values, as per torchvisions center_crop documentation .This is done silently currently, I think we should either explicitly throw an exception for this case of input size smaller than requested output size as I don't see it as a case we want to support/part of the original intention of this function or at least throw a warning so it's more obvious this is happening, keen to hear thoughts on this, thanks! @peterdudfield @dfulu @AUdaltsova
The text was updated successfully, but these errors were encountered: