Error [defaultAllocator.cpp::deallocate::35] Error Code 1: Cuda Runtime (invalid argument)

[ART97]

Hi
I have converted yolov5s.pt file into yolov5s.wts and then build the engine file yolov5s.engine. I have modified some parts of the code from here Jetson/L4T/TRT Customized Example - eLinux.org and some code taken from this repo tensorrtx/yolov5_det_trt.py at master · wang-xinyu/tensorrtx · GitHub . Everything is working fine but when I close the window by pressing Esc, it closes with error msg:

[03/30/2023-19:08:59] [TRT] [E] 1: [defaultAllocator.cpp::deallocate::35] Error Code 1: Cuda Runtime (invalid argument)

Tensor rt version: 8.2.1.8
Board Type: Jetson Nano
Cuda: 10.2.300
Jetpack: 4.6.1

I have uploaded my project here: test.zip - Google Drive
with all the files. Please if anyone can suggest what is the issue? Thanks

[AastaLLL]

Hi,

This is related to the destructor of TensorRT.
Could you try to unset the engine variable to see if it helps?

engine = []

Thanks.

[ART97]

Hi @AastaLLL
Where exactly I have to put this in code I shared?
I created another function in the class YoloV5TRT

def Destroy(self):
   self.engine = []

and called this outside of while display.IsStreaming(): so that once application closes, self.engine = [] is called but still getting the same error.

Thanks

[ART97]

Hi, @AastaLLL is there any update on this, please? Thanks

[ART97]

Solved it by moving the below variable outside of the class

host_inputs
cuda_inputs
host_outputs
cuda_outputs
bindings

[AastaLLL]

Hi,

Sorry for the late reply.
Thanks for the update and good to know it works now.