search

triple-Mu / YOLOv8-TensorRT

YOLOv8 using TensorRT accelerate !
MIT License
1.32k stars 228 forks source link

index out of range #225

Open WeisonWEileen opened 3 months ago

WeisonWEileen commented 3 months ago

image

when I input my onnx file to build.py, it turn out that that is a index out error. Could you please help me how to debug?

triple-Mu commented 3 months ago

image

when I input my onnx file to build.py, it turn out that that is a index out error. Could you please help me how to debug?

The process is like this:

  1. Using export_det.py for exporting onnx.
  2. Using build.py build engine by the exported onnx
  3. Using infer_det.py for object detection.

It seems that your onnx does not came from export_det.py.

WeisonWEileen commented 3 months ago

Thanks，you are true. And there are problems puzzle me:

  1. Does these red error matters, which comes when I export onnx from pt?
  2. when I run /bin/trtexec to export engine from onnx, it cost me almost 10 min, are there anyway to accelarate? my platform is jetson orin nx 16GB image
triple-Mu commented 3 months ago

Thanks，you are true. And there are problems puzzle me:

  1. Does these red error matters, which comes when I export onnx from pt?
  2. when I run /bin/trtexec to export engine from onnx, it cost me almost 10 min, are there anyway to accelarate? my platform is jetson orin nx 16GB image

The red error comes from onnxruntime. It doesn't make any sense. Do not use --fp16 flag. If you want to go further, you can try build-tensorrt-by-api