search

deepcam-cn / yolov5-face

YOLO5Face: Why Reinventing a Face Detector (https://arxiv.org/abs/2105.12931) ECCV Workshops 2022)
GNU General Public License v3.0
2.12k stars 503 forks source link

TensorRT questions #109

Open Jim0530 opened 2 years ago

Jim0530 commented 2 years ago

Using yolov5n-0.5.pt on Jetson Nano and Get about 5 fps And I try to use tensorRT So I first using models/export.py and torch2tensorrt/main.py to do inference And I get some questions

  1. When building engine it took about 5 or 6 minutes?
  2. What can I do for the .trt file? Can I load a model via .trt
  3. It took 70s to inference a picture

Is there any implements details that I missed? Thank u sincerely if u can help me out!

bobo0810 commented 2 years ago

https://github.com/deepcam-cn/yolov5-face/issues/76#issuecomment-939733533

https://github.com/deepcam-cn/yolov5-face/issues/63

Jim0530 commented 2 years ago

I saw ur update this morning How should I set the Explicit Batch? Thank you

bobo0810 commented 2 years ago

I saw ur update this morning How should I set the Explicit Batch? Thank you

https://github.com/deepcam-cn/yolov5-face/blob/ea3ed6a7b548a4786ad8cf7fae333ea5c342cd31/export.py#L27

Jim0530 commented 2 years ago

Thanks a lot!

Jim0530 commented 2 years ago

I got an error Attribute error "tensorrt.tensorrt.Builder" has no attribute "build serialized network" I guess that's due to the different version of tensorrt am I right? (I am using Tensorrt 7.1.3 now)

bobo0810 commented 2 years ago

I got an error Attribute error "tensorrt.tensorrt.Builder" has no attribute "build serialized network" I guess that's due to the different version of tensorrt am I right? (I am using Tensorrt 7.1.3 now)

try 8.2

Jim0530 commented 2 years ago

Works for me thanks

Wanghe1997 commented 2 years ago

Using yolov5n-0.5.pt on Jetson Nano and Get about 5 fps And I try to use tensorRT So I first using models/export.py and torch2tensorrt/main.py to do inference And I get some questions

  1. When building engine it took about 5 or 6 minutes?
  2. What can I do for the .trt file? Can I load a model via .trt
  3. It took 70s to inference a picture

Is there any implements details that I missed? Thank u sincerely if u can help me out!

Hello, regarding the yolov5-face project. Did you finally succeed in reasoning about the tensorrt-accelerated .engine model on Jetsonnano? If so, how was it implemented? Thank you!