Deploy TRT Object Detection model (Mobilenetv2) with Deepstream Error:"Failed to parse bboxes"

Accelerated Computing Intelligent Video Analytics DeepStream SDK
3

Hi,

The error occurs since the customized parser information doesn’t be added to the configure file.

Without this information, Deepstream will parse the mobilenet output with a standard detection parser.
And will expect a ‘converge’ and ‘bboxes’ output layer, which is not available in your model.

To solve this, please add the following information into the config_infer_primary_test.txt, and try it again.

[property]
..
force-implicit-batch-dim=1
parse-bbox-func-name=NvDsInferParseCustomSSD
custom-lib-path=nvdsinfer_custom_impl_ssd/libnvdsinfer_custom_impl_ssd.so

Here is a related comment for your reference:

Thanks.