Shape input layer issue about channel color

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

• Hardware Platform Jetson
• DeepStream 5.0
• JetPack Version 4.4
• TensorRT Version 7.1

• Issue Type:
I had convert keras model to onnx model and buid engine successful with layer input size:110x47x1 (channel last). When i run deepstream app with this model and engine file, the result is an error:

GRAY input format specified but network input channels is not 1

• i have change property:
model-color-format=2

So how can add a transpose (CHW → HWC) operation on deepstream app ?

3

Hi,

Currently, Deepstream doesn’t support the input order configuration for ONNX model.
Please convert the model into NCHW input when generating the ONNX file.

Thanks.

1 Like
4

Hi, HCHW or NCHW
Can you give me an example?
thank @AastaLLL

5

Hi,

Sorry for the typo. It’s NCHW.
You can find some discussion for ONNX NHWC and NCHW conversion here:
https://github.com/onnx/onnx/issues/369

Thanks.

6

Thanks you