Yolo-plugins make failure on Xavier

Autonomous Machines Jetson & Embedded Systems Jetson AGX Xavier
1

Hi,

I tried yolo-plguin on P4 (https://github.com/vat-nvidia/deepstream-plugins), and it works pretty well. But when it switched to the Xavier platform, I got a make error:

trt_utils.h:83:22: error: ‘nvinfer1::DimsHW YoloTinyMaxpoolPaddingFormula::compute(nvinfer1::DimsHW, nvinfer1::DimsHW, nvinfer1::DimsHW, nvinfer1::DimsHW, nvinfer1::DimsHW, const char*)’ marked ‘override’, but does not override nvinfer1::DimsHW compute(nvinfer1::DimsHW inputDims, nvinfer1::DimsHW kernelSize,

Any suggestion?
thx.

2

Hi,

yolo-plugin requires deepstream SDK which is not available on Xavier.
For now, please use the trt-yolo-app instead.

Thanks.

3

Thx for reply.
The error occurs when I compile the trt-yolo-app.

4

Have you solved the problem?

5

No. The current repo is not verified on Xaver, and the specific version for Xavier will be released in coming months.

1 Like
6

Hi,

We usually test it on TX2, which is using JetPack3.3. Not sure if any issue with JetPack 4.1 for Xavier.
We are trying to reproduce this internally. Will update our investigation with you.

Thanks.

7

Hi,

We can also reproduce this issue in our environment and are working on it.
Will update information with you later.

Thanks.

8

Hi,

Error is caused by some API update in TensorRT 5.0.

Please apply this changes to fix the conflict:

diff --git a/sources/lib/trt_utils.h b/sources/lib/trt_utils.h
index cec1612..c04174a 100644
--- a/sources/lib/trt_utils.h
+++ b/sources/lib/trt_utils.h
@@ -82,7 +82,7 @@ private:
 
     nvinfer1::DimsHW compute(nvinfer1::DimsHW inputDims, nvinfer1::DimsHW kernelSize,
                              nvinfer1::DimsHW stride, nvinfer1::DimsHW padding,
-                             nvinfer1::DimsHW dilation, const char* layerName) override
+                             nvinfer1::DimsHW dilation, const char* layerName) const override
     {
         assert(inputDims.d[0] == inputDims.d[1]);
         assert(kernelSize.d[0] == kernelSize.d[1]);

Thanks.