Capture a video from the camera on a platform (etc. Jetson Xavier) and the compressed and encoded frames are transmitted to another paltform (etc. Jetson TX), where the frames are decoded and reconstructed in real time.

1. python 3.6.9
2. pycuda 2022.1
3. tensorrt 7.1.3.4
4. opencv 4.1.1
5. grpc 1.48.2
6. compressai 1.2.4
7. torch 1.9.0
8. torchvision 1.10.0
9. numpy 1.19.5

1. make sure the ip address works.
2. make sure .trt files match the platform.
3. Folders int8/flt_onnx generate .onnx models with executing ImplicitQ/flt_*.py directly.
4. Folders Int8LIC_*2/127 build Int8 mode TensorRT engines (.trt files) from .onnx models, where 2 means /2 and folder 127 means -127; build_*_trt.py builds Float32 mode TensorRT engines from .onnx models.
5. The input size is fixed, which is [1, 3, 256, 256].
6. TX2 platform can not support Int8 mode of TensorRT.
7. flt_decoder_cmp.py & flt_encoder_cmp.py: floating-point pretrained models with flt_*.onnx & flt_*_TX/XA.trt.
8. decoder_cmp.py & encoder_cmp.py: int8 quantized models with Int8LIC_*.onnx & Int8LIC_*_TX/XA.trt.
9. flt_decoder_cmp.py & decoder_cmp.py use D & pD, while flt_encoder_cmp.py & encoder_cmp.py use E & pD.

python3 flt_decoder_cmp.py --ip 192.168.1.188:50051

python3 flt_encoder_cmp.py --ip 192.168.1.188:50051

• The ip addresses on two machines should keep the same.