accuracy

[duong0411]

Why do my accuracy is negative ?

[triple-Mu]

Why do my accuracy is negative ?

It may be caused by fp16 engine in TensorRT<=8.2.4

[duong0411]

im convert with tensorrt in jetson nano. I will try 2 methods for export onnx with end2end model nms:

• methods1: im using python api on your github. About conf, threshold when convert onnx and tensorrt, can i set same config but setup different config with onnx: conf:0.5 and thres: 0,65 and tensorrt: conf: 0.6 thres 0.65 and inference model my acc is negative, size model engine bigger model onnx. I dont understand
• method2: im using trtexec tool but it is fail.

[duong0411]

Using trtexec tools for export engine

/usr/src/tensorrt/bin/trtexec
--onnx=yolov8s.onnx
--saveEngine=yolov8s.engine
when i export with --fp16, accuracy is negative ?

[triple-Mu]

Using trtexec tools for export engine

/usr/src/tensorrt/bin/trtexec --onnx=yolov8s.onnx --saveEngine=yolov8s.engine when i export with --fp16, accuracy is negative ?

Same issue in NVIDIA/TensorRT#1758