i don’t want to touch primary-gie for human detection, i want to get secondary results of human just such age, gender detection , secondary-gie for cars.
so how do i replace the secondary -gie, should i modify any files in ds sdk?
and should i transform caffe model into tensorRT version(i don’t really know much about tensorRT)
so if you guys can explain this implicitly and maybe attach a link about how to transform caffe model with tensorRt requirements ,
i will be super grateful .
thanks sooo much .
Deepstream 4.0 meets your requirement.
Please read the documents, read the code, run samples to clear deepstream design.
You can deploy your caffe model by tensorRT Firstly. TensorRT support caffe model directly. If there is someone layer tensorRT does not support, you need to implement tensorRT iplugin.
Replace deepstream pgie, sgie by your model, just config should be changed.
Refer to sources/libs/nvdsinfer_customparser to define your own output parser if the default parser is right for you.
tensorRT iplugin sample is here: sources/objectDetector_FasterRCNN/nvdsinfer_custom_impl_fasterRCNN