Size of input tensor in ReformatImage function

[dattv]

Hi all.

I'm wondering that input resolution of yolov5 is 3x672x672, but why in

cuDLA-samples/src/matx_reformat/matx_reformat.cu

Line 146 in a2d645b

matx::tensor_t<matx::matxFp16, 5> mInput1 = matx::make_tensor<matx::matxFp16>((matx::matxFp16*)src[0], {1, 16, 672, 672, 1});

, the shape of mInput1 tensor was allocated as (1x16x672x672). (in case of using fp16)
anybody known why???
@zerollzeng, @mchi-zg

[zerollzeng]

Because DLA takes chw16 inputs, which means we need to pad the input to NC/16HW16 first.

[dattv]

Because DLA takes chw16 inputs, which means we need to pad the input to NC/16HW16 first.

@zerollzeng Thank you for your feedback!,
however there are some extra information I want to inquires,

• Did you means 16 in chw16 is 16 Bit?
• Because the image included 3 channel (RGB) so i thought the mInput1 tensor must be something like 1x3x672x672*sizeof(half)?
• Is there any different between memory size of cuda variables and dla variables?

[zerollzeng]

No, the 16 means the C channel is padded to 16, So you actually have a 1x16x672x672 tensor, while only the first 3 channels represent the image.

[lynettez]

closing since no activity for several month, thanks all!