===Optimize model ===
/usr/src/tensorrt/bin/trtexec --onnx=pallet_yolov4_v1.onnx --buildOnly --saveEngine=pallet_yolov4_416.engine

===Inference by video ===
gst-launch-1.0 filesrc location=pallet.mp4 ! decodebin ! nvvidconv ! videoconvert ! adrt model=pallet_yolov4_416.engine scale=0.004 mean="0 0 0" rgbconv=True ! adtrans_yolo class-num=1 anchor=anchor="(12,16),(19,36),(40,28),(36,75),(76,55),(72,146),(142,110),(192,243),(459,401)" threshold=0.3 label=pallet_yolov4_v1.names use-sigmoid=True ! admetadrawer ! videoconvert ! xvimagesink

===Inference and save video ===
gst-launch-1.0 filesrc location=pallet.mp4 ! decodebin ! nvvidconv ! videoconvert ! adrt model=pallet_yolov4_416.engine scale=0.004 mean="0 0 0" rgbconv=True ! adtrans_yolo class-num=1 anchor=anchor="(12,16),(19,36),(40,28),(36,75),(76,55),(72,146),(142,110),(192,243),(459,401)" threshold=0.3 label=pallet_yolov4_v1.names use-sigmoid=True ! admetadrawer ! videoconvert ! omxh264enc bitrate=4000000 ! h264parse ! qtmux ! filesink location=filename.h264