lobanov-m commented on a change in pull request #13226: [MXNet-1211] Factor and
"Like" modes in BilinearResize2D operator
URL: https://github.com/apache/incubator-mxnet/pull/13226#discussion_r236169303
##########
File path: src/operator/contrib/bilinear_resize-inl.h
##########
@@ -105,8 +131,11 @@ inline void BilinearSampleOpBackward(const
nnvm::NodeAttrs& attrs,
const std::vector<TBlob> &inputs,
const std::vector<OpReqType> &req,
const std::vector<TBlob> &outputs) {
- CHECK_EQ(inputs.size(), 1U);
- CHECK_EQ(outputs.size(), 1U);
+ const BilinearSampleParam& param =
nnvm::get<BilinearSampleParam>(attrs.parsed);
+ CHECK_EQ(inputs.size(), 1);
+ bool modeLike = param.mode == bilinear_resize::like;
+ size_t expected = modeLike ? 2 : 1;
+ CHECK_EQ(outputs.size(), expected);
Review comment:
This is in the backward function. In "like" mod we pass two tensors to
resize one to the size of second, so the backward function should return
gradients to both tensors. Actually the second tensor, from which we get result
size, should get zero gradients from the output of this operator, because it is
needed only to get it's shape. This is realized in cpp files.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
