Serialization Error in verifyHeader: 0 (Version tag does not match) while running deepstream-app with YOLOv4

[MGh]

Please provide complete information as applicable to your setup.

• Hardware Platform (Jetson / GPU) Tesla T4
• DeepStream Version 5.0
• TensorRT Version 8.0.0-1
• NVIDIA GPU Driver Version (valid for GPU only) 11.3

I am following the steps given in GitHub - NVIDIA-AI-IOT/yolov4_deepstream to create a tensorrt engine file, and then I use it to run deepstream-app. However, I get the following error:

Based on this topic and a few other topics, this is due to a mismatch between tensorrt version used for creating the engine and the one for deserializing it. However, I am creating the engine on the same machine and I assume that the same version of tensorrt should be used for both task.
Do you know what can possibly cause this problem?

Thanks

[Amycao]

Did you build the engine and deploy it to deepstream in the same environments? did you run the ds app in docker or in the host?

[MGh]

@Amycao Yes, I built the engine and deployed it into deepstream in the same environment. Also, the ds is running in the host as well.

[Amycao]

Did you build the engine follow this Tianxiaomo/pytorch-YOLOv4: PyTorch ,ONNX and TensorRT implementation of YOLOv4 (github.com)? or this, yolov4_deepstream/tensorrt_yolov4 at master · NVIDIA-AI-IOT/yolov4_deepstream (github.com)

[MGh]

Hi @Amycao

I am following the first one, i.e., GitHub - Tianxiaomo/pytorch-YOLOv4: PyTorch ,ONNX and TensorRT implementation of YOLOv4

[Amycao]

There is no update from you for a period, assuming this is not an issue anymore.
Hence we are closing this topic. If need further support, please open a new one.
Thanks

• TensorRT Version 8.0.0-1

From the description, you used TRT 8.0.0, for deepstream 5.0, you should use TensorRT 7.0.X