Resize layer for DLA

Autonomous Machines Jetson & Embedded Systems Jetson Xavier NX
#1

Hi,
Attached is a small test. For some reason the Resize layer does not run on the DLA, any ideas?
I’m using Jetpack 4.4 and the resize layer should have gone to the DLA.

Output:
WARNING: Default DLA is enabled but layer (Unnamed Layer* 0) [Resize] is not supported on DLA, falling back to GPU.
WARNING: setDefaultDeviceType was called but no layer in the network can run on DLA.
UNKNOWN: Applying generic optimizations to the graph for inference.
INFO:
INFO: --------------- Layers running on DLA:
INFO:
INFO: --------------- Layers running on GPU:
INFO: (Unnamed Layer* 0) [Resize],

Test.cu (2.9 KB)

[Compiling using:
nvcc --Wno-deprecated-declarations Test.cu -lnvinfer -lrvinference -lnvToolsExt -lnvonnxparser -o Test
]

#3

Hi,

Since DLA is a hardware-based inference engine, not all the TensorRT layers can run on DLA.
Please check the DLA support matrix below. The resize layer is not on the list.

Thanks.

#4

Hi @AastaLLL,
Thanks, I am probably missing something here… The link below (TRT 7.2.2) shows it is supported…
Which is correct? 7.2.2 is still not available for Jetson? where do I see which version I have?

[UPDATE] - I see that 7.1.3 is in Jetpack 4.4.1 and so I’ll have to wait for 7.2.2 to come out?

Resize layer

  • Only Nearest Neighbor resize mode is supported.
  • The number of scales has to be exactly 4.
  • The first two elements in scales need to be exactly 1 (for unchanged batch and channel dimensions).
  • The last two elements in scales, representing the scale values along height and width dimensions, respectively, need to be integer values in the range of [1, 32].

Thanks
Eyal

#5

Hi,

Yes. The resize layer with DLA is new for TensorRT 7.2.
Currently, the latest TensorRT version for Jetson is still TensorRT 7.1.

You will need to wait for the new release to get the layer support.
For now, you can use GPU to inference it as a temporary workaround.

Thanks.

1 Like