Invalid configuration argument (error 9)

[hmedicharla]

Hi,
I am getting invalid configuration error and I am clueless about its cause and way to debug this. Please help me debug this issue.

I am using the dusty_nv/jetson-inference as the code base but instead of feeding the default models as input to this code, I am using a customised SSD model. And attached is the log that I get.
log_custom_ssdmodel.txt (285.3 KB)

[image] loaded ‘images/dog_123.jpg’ (500x375, 3 channels)
[cuda] cudaAllocMapped 562500 bytes, CPU 0x21ecab000 GPU 0x21ecab000
[TRT] ExecuteV2 TensorRT context on device GPU
[cuda] invalid configuration argument (error 9) (hex 0x09)
[cuda] /home/deepedge1/hema/dusty_nv_t1/jetson_inference_2/9thDec2022_default/c/detectNet.cpp:987
[TRT] detectNet::Detect() – failed to render overlay
1968 objects detected

I am using the image of a dog (attached) for inference and as per the log, looks the detection is not being done.

1968 objects detected
detected obj 0 class #3 (car) confidence=0.909668
bounding box 0 (150.110321, 0.002790) (0.004776, 0.000063) w=-150.105545 h=-0.002727
detected obj 1 class #3 (car) confidence=0.920898
bounding box 1 (89.298164, 0.002736) (0.000057, 0.000102) w=-89.298111 h=-0.002634
detected obj 2 class #2 (bicycle) confidence=0.510742
bounding box 2 (0.000000, 0.000000) (0.000000, 0.000000) w=0.000000 h=0.000000
detected obj 3 class #2 (bicycle) confidence=0.519043
bounding box 3 (0.000000, 0.000000) (0.000000, 0.000000) w=0.000000 h=0.000000

Thanks,
Hema

[AastaLLL]

Hi,

Does your customized model use the same input/output name as the default SSD model?
If not, please run it with --input-blob=[] and --output-blob=[] flag.

You can find an example below

detectnet on custom model (ssd-mobilenet-v2) got error cuTensor Error in executeCutensor: 7 (Internal cuTensor reformat failed)

Thanks.