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_r236167407
##########
File path: src/operator/contrib/bilinear_resize-inl.h
##########
@@ -44,17 +44,39 @@
#include "../mxnet_op.h"
#include "../mshadow_op.h"
+namespace bilinear_resize {
+enum BilinearResizeOpMode{size, scale, odd_scale, like, to_even};
+} // namespace bilinear_resize
+
+
namespace mxnet {
namespace op {
struct BilinearSampleParam : public dmlc::Parameter<BilinearSampleParam> {
- int height;
- int width;
+ float height;
+ float width;
+ int mode;
DMLC_DECLARE_PARAMETER(BilinearSampleParam) {
- DMLC_DECLARE_FIELD(height).set_range(1, 10000)
- .describe("output height (required)");
- DMLC_DECLARE_FIELD(width).set_range(1, 10000)
- .describe("output width (required)");
+ DMLC_DECLARE_FIELD(height).set_range(0, 10000)
+ .set_default(-1)
+ .describe(R"(output height if mode is "size" or scale on which input
height should be multiplied if mode is
+ "scale" or "odd_scale")");
+ DMLC_DECLARE_FIELD(width).set_range(0, 10000)
+ .set_default(-1)
+ .describe(R"(output width if mode is "size" or scale on which input width
should be multiplied if mode is
+ "scale" or "odd_scale")");
+ DMLC_DECLARE_FIELD(mode)
+ .add_enum("size", bilinear_resize::size)
+ .add_enum("scale", bilinear_resize::scale)
+ .add_enum("odd_scale", bilinear_resize::odd_scale)
+ .add_enum("like", bilinear_resize::like)
+ .add_enum("to_even", bilinear_resize::to_even)
+ .set_default(bilinear_resize::size)
+ .describe(R"(resizing mode. "size" - resize to distinct size; "scale" -
original height and width are multiplied
Review comment:
I think any. If we have original tensor with shape (1, 3, 10000, 10000) and
resize it with height=0.001 and width=0.001 in "scale" mode we will get tensor
with shape (1, 3, 10, 10). And back we can resize the result tensor in scale
mode with height=100 and width=100 and get tensor of original size. I'll try to
clarify this in description.
----------------------------------------------------------------
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
