Semantic segmentation on PX2 with a model pretrained on Cityscapes

Autonomous Vehicles DRIVE PX 2 DRIVE - Linux
1

Dear all,
I would like to do semantic segmentation (i.e. run inference) with a neural network trained on Cityscapes such as MobileNet-v3 or Xception_71 [1]. I use PX2 (AutoChaffeur, Ubuntu 16.04, AArch64), so I can only use C++ API rather than Python API, as the latter is just not available. Obviously, I would also need to use TensorRT for my task. I had been wondering if somebody could suggest some particular tutorials I should follow; I know that there’s quite a lot of them on NVIDIA website, but it is precisely this abundance of information that makes me feel a bit at loss… Thanks a lot in advance!

[1] https://github.com/tensorflow/models/blob/master/research/deeplab/g3doc/model_zoo.md

2

Dear shader2020,
We have provided DLI lab at GTC on this earlier but it is not available in public domain. If you already have a trained model, you can take a look at TensorRT samples on optimizing your model furthur. Let us know if you face any issues. Also, note that the last release on DRIVE PX 2 has TensorRT 4.0 and all the releases now are targetted towards to DRIVE AGX platform. If possible consider upgrading to DRIVE AGX platform to get latest SW stack.

3

Dear SivaRamaKrishna,
Thanks a lot!