Problem in facing in converting custom yolo weights into TensorRT

Autonomous Machines Jetson & Embedded Systems Jetson TX2
1

Good evening
I am working on TensorRT models. While converting custom yolo model into TensorRT, it is showing value error. What is that error and how to solve that error. I am enclosing screenshot of the error. Please help me to sort out the error.

IMG20210415110344_00
IMG20210415110344_004608×3456 2.94 MB

3

Hi,

YOLOv4 requires an extra support in the [route] layer.
You can find the detailed here.

If Deepstream is an optional for you, below is an example to run YOLOv4 Tiny with Deepstream objectDetector_Yolo sample:
https://elinux.org/Jetson/L4T/TRT_Customized_Example

Thanks.

4

I got that error after using the custom dataset. please tell me how to solve that error. thanks in advance.

5

Hi,

Please noted the default /usr/src/tensorrt/samples/python/yolov3_onnx/ sample is for YOLOv3.
It needs some update for the YOLOv4 model, which is basically for the [route] layer.

We do this for the deeptream sample before (please check the link above on April 16) .
You can go to the sample and apply the patch to generate the plan file.
Once you got the file, you can copy it back to TensorRT sample and run the following.

Thanks