Skip to main content

YOLO Usage

Notes
  • EC YOLOv8 requires EC Android 10.15.0+
  • EC Android integrates Tencent/ncnn optimized for mobile devices
  • ncnn supports YOLOv5–YOLOv8; this guide uses YOLOv8 for training, inference, and detection
  • YOLOv8 official site https://docs.ultralytics.com/modes/train/
  • YOLO Demo download: Resources, cloud drive YOLO Resources → YOLODemo.zip

Install & Train Model​

Install Software​

  • After installing Anaconda, create a virtual environment
  • Click Create, name it yolotest (or other name), Python 3.8.19, Create

  • Click green triangle on yolotest → Open Terminal

  • Terminal prompt should show (yolotest)

Configure Environment​

  • Install Python packages in terminal:
 pip config set global.index-url https://pypi.tuna.tsinghua.edu.cn/simple/
 pip config set install.trusted-host pypi.tuna.tsinghua.edu.cn
 # yolo 0.3.1
 # Latest: pip install yolo
 pip install yolo==0.3.1
 # ultralytics 8.2.79
 # Latest: pip install ultralytics
 pip install ultralytics==8.2.79
 pip install ncnn==1.0.20240410
 pip install labelimg
 # Finally: pip list
  • yolo, ultralytics installed

  • Run yolo in terminal to verify

  • Run labelimg to open annotation tool

Create Training Directory​

  • Create folder e.g. yolotrain on E: drive
  • Under yolotrain: labels/, images/; under images/: test, train, val; under labels/: train, val
  • Structure:

  • Put same training images in images/train, images/val, and images/test (copy files)
  • labels/train and labels/val hold labelimg annotation files

Annotate Data​

  • Put images in images/train, copy to images/val and images/test
  • Run labelimg from configured terminal
  • Open Dir → images/train (e.g. E:/yolotest/images/train)
  • Change Save Dir → labels/train (e.g. E:/yolotest/labels/train)
  • Set format to YOLO via Save format button

  • Annotate
  • Example: Like and Comment buttons on a short-video app
  • Right-click → Create RectBox or click left Create RectBox
  • Draw box on Like area, class name aixin, OK — used in API calls later
  • Draw box on Comment, class pinglun, OK
  • Save, Next Image for next image
  • labels/train gets classes.txt and per-image .txt files
  • Copy labels/train to labels/val for validation

Train Model​

  • In yolotrain, create aixin.yaml:
 path: E:/yolotrain
 train: images/train
 val: images/val
 test: images/test
 nc: 2
 names: ["aixin","pinglun"]

  • Parameters:

  • path: training root (adjust drive/path)

  • train, val, test: image folders relative to path

  • nc: number of classes (2 here)

  • names: class names from labelimg — order matters

  • Train in cmd: e:/ then cd yolotrain

  • Training command (pick one):

 yolo detect train data=e:/yolotrain/aixin.yaml model=e:/yolotrain/yolov8s.pt imgsz=640
 yolo detect train data=e:/yolotrain/aixin.yaml model=e:/yolotrain/yolov8s.pt epochs=100 imgsz=640
  • Training in progress
  • Done — Results saved path varies; example runs/detect/train under E:/yolotrain
  • Find best.pt and annotated result images in output folder

Validate Model​

  • Validate on images:
 yolo detect val data=e:/yolotrain/aixin.yaml model=e:/yolotrain/runs/detect/train/weights/best.pt
  • Check Results saved path for annotated images

Export ONNX Model​

  • Export trained .pt to ONNX for EC API
 yolo export model=e:/yolotrain/runs/detect/train/weights/best.pt format=onnx
  • export success shows path — copy .onnx to phone /sdcard/ (or other path)
  • Initialize with yolov8Api.newYolov8Onxx; other API calls are the same

Export NCNN Model​

  • Export to ncnn for EC
 yolo export model=e:/yolotrain/runs/detect/train/weights/best.pt format=ncnn
  • Downloads pnnx — use demo pnnx-20240819-windows.zip in e:/yolotrain/ if download fails
  • Match filename in error screenshot or download manually

  • Output has model.ncnn.param and model.ncnn.bin — copy both to phone sdcard

  • Initialize with yolov8Api.newYolov8 for ncnn

API Usage​

GPU​

CONDA Virtual Environment Error​

  • Cannot select Python version when creating env — often network; switch to Tsinghua mirror
    1. Click channels

    1. Remove default channels

    1. Add Tsinghua channels (Enter after each)
 https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/
 https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/

    1. Click Update channels

Arial.ttf Download Error​

  • Download Arial.ttf
  • Place in C:\Users\Administrator\AppData\Roaming\Ultralytics as Arial.ttf (Administrator = your Windows username)