I am trying to use the 8 bit quantized version of peoplenet (PeopleNet | NVIDIA NGC), but in my deepstream pipeline, when specifying the network mode as int8 , I get a warning about the missing calibration file - WARNING: INT8 calibration file not specified. Trying FP16 mode.
This file is not provided with any of the peoplenet models, and the etlt file I am providing is named
resnet34_peoplenet_int8.etlt, as one would expect from the quantized model.
So is the engine file that is beeing built (and subsequently used) actually an int8 model or an fp16 model?
And where can the calibration files for this model be found?