YOLOX is a high-performance anchor-free YOLO, exceeding yolov3~v5 with MegEngine, ONNX, TensorRT, ncnn, and OpenVINO supported. Documentation: https://yolox.readthedocs.io/
Apache License 2.0
9.39k
stars
2.2k
forks
source link
TensorRT-int8 model is NOT faster than TensorRT-float16 model #1001
I trained yolox-m model and convert it to TensorRT-int8 model with demo/trt.py.
Difference from original trt.py is torch2trt's args , float16_mode=false and int8_mode=True, and calibration data [data].
I trained yolox-m model and convert it to TensorRT-int8 model with demo/trt.py. Difference from original trt.py is torch2trt's args ,
float16_mode=false
andint8_mode=True
, and calibration data[data]
.Infer times in tools/demo.py are like bellow.( Aws EC2 p3.2xlarge with Nvidia Tesla V100)
Other models( yolox-s, yolox-l) seems to be the same.
When using YOLOX, Int8 model is not faster than float model? Or I make some mistakes?