DeepStream Version: 6.0 (as per your previous context)
JetPack Version: 4.6.5
TensorRT Version: 8.2.1.8
CUDA Version: 10.2.300
cuDNN Version: 8.2.1.32
I developed an interface using your DEEPSTREAM that uses the RESNET10 model to identify: cars, people, bicycles, and traffic lights. In addition, I added to PIPELINE another processing that classifies vehicles by category. (This is also the RESNET10 model).
Now I want to integrate a model that detects lanes on the road and was trained by data CULANE. This model is (ONNX,TRT).
you can use this pipeline “…pgie(detection)-> sgie1(classfication->sgie2(detection)->…)”. please refer to sample /opt/nvidia/deepstream/deepstream/sources/apps/sample_apps/deepstream-test2/. the pgie detects vehicles, the sgie classifies vehicles.
if using Onnx models, please refer to this sample code.
I can’t understand.
How can I combine models of different types in the same PIPELINE??
“you can use this pipeline “…pgie(detection)-> sgie1(classfication->sgie2(detection)->…)”. please refer to sample /opt/nvidia/deepstream/deepstream/sources/apps/sample_apps/deepstream-test2/. the pgie detects vehicles, the sgie classifies vehicles.”
I already know that!
Primary mode means operating on full frames. Secondary mode means operating on objects added in the meta by upstream components. Since the two detection models have no dependency, you can use " pgie(detection) → sgie(classification) → pgie(detection)" pipeline. especially please set process-mode of both detection nvinfer to 1, which means process_full_frame=true.
There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks