Sign inSign up

lianshufeng/vframeengine

By lianshufeng

Updated over 1 year ago

Image
0

1.6K

lianshufeng/vframeengine repository overview

Features

  • yolo11 detect
  • yolo11 segment
  • yolo11 pose
  • deepface
  • Qwen2.5-VL
  • Support multi-resolution video output
  • Support API service
  • Support Async Inference

Install

  • gpu

https://www.nvidia.cn/geforce/drivers/
https://developer.nvidia.com/cuda-downloads
  • python
python 3.11 
  • yolo
https://docs.ultralytics.com/zh#where-to-start
  • script
install.cmd
  • tensorrt
# https://developer.nvidia.com/tensorrt/download/10x

# pip  uninstall tensorrt -y

pip install tensorrt-10.7.0-cp311-none-win_amd64.whl
  • tensorflow



> https://www.tensorflow.org/install/gpu
NVIDIA® GPU 驱动程序 - CUDA® 11.2 要求 450.80.02 或更高版本。
CUDA® 工具包:TensorFlow 支持 CUDA® 11.2(TensorFlow 2.5.0 及更高版本)
CUDA® 工具包附带的 CUPTI。
cuDNN SDK 8.1.0 cuDNN 版本。
(可选)TensorRT 6.0,可缩短用某些模型进行推断的延迟时间并提高吞吐量。

train


# 设置工作路径
yolo settings datasets_dir=/work


#开始训练 device=cpu | 0,1 (多卡)
yolo detect train data=datasets/yolo/dataset.yaml model=yolo11n.pt epochs=100 imgsz=640 

#恢复训练 
yolo train resume model=runs/detect/train/weights/last.pt

#继续训练
yolo detect train data=datasets/yolo/dataset.yaml model=runs/detect/train/weights/last.pt epochs=500

Export

  • engine
# 导出 TensorRT, 注: windows 和 linux 导出的不能跨平台运行
yolo export model=yolo11n.pt format=engin

常见问题

  • Qwen2.5-VL RuntimeError: cutlassF: no kernel found to launch!
# 多卡训练时,需要设置环境变量 CUDA_VISIBLE_DEVICES=0,1,2,3
# win
set CUDA_VISIBLE_DEVICES=0
# linux
export  CUDA_VISIBLE_DEVICES=0

Tag summary

Content type

Image

Digest

sha256:3a72dd2c9

Size

16.2 GB

Last updated

over 1 year ago

docker pull lianshufeng/vframeengine