How to upsamle a layer into the shape of another layer in the symbol api?

zhoulukuan · July 19, 2019, 10:25am

In pytorch, we can use:

h, w = x.size(2), x.size(3)
y = F.upsample(y, size=(h, w), mode='bilinear')

to upsample and make the layer y have the same shape with layer x. How to implement this function in mxnet symbol api? BilinearResize2D needs to be provided the output size and cannot dynamically change the size after upsamle. So I am confused how to to.

QueensGambit · July 19, 2019, 3:11pm

Hi @zhoulukuan,
I created an UpsampleBlock in Gluon in the past.

The syntax is very similar:

F.UpSampling(x, scale=self.scale, sample_type='bilinear')

https://beta.mxnet.io/api/ndarray/_autogen/mxnet.ndarray.UpSampling.html

github.com

QueensGambit/CrazyAra/blob/master/DeepCrazyhouse/src/domain/neural_net/architectures/rise_builder_util.py#L141


        out = self.bn0(out)
        out = self.act0(out)


        # apply squeeze excitation
        if self.use_se:
            out = self.se0(out)


        return out




class _UpsampleBlock(HybridBlock):
    def __init__(self, name, scale=2, sample_type="nearest"):
        super().__init__(prefix=name)
        self.scale = scale
        self.sample_type = sample_type


    def hybrid_forward(self, F, x):
        """
        Compute forward pass


        :param F: Handle

Greetings,
~QueensGambit

Topic		Replies	Views
How can I use Interpolate without learnable parameters mxnet	2	721	October 23, 2019
Looking for 2 functions in gluon	4	953	April 11, 2018
Using UpSampling in HybridBlock Gluon	1	1409	October 4, 2018
mx.nd.contrib.BilinearResize2D (Inside HybridBlock) Gluon	3	1621	July 24, 2018
Get HybridBlock layer shape on runtime Gluon	7	3490	May 17, 2018

How to upsamle a layer into the shape of another layer in the symbol api?

Related Topics