###### tags: `yolov5` # Yolov5 使用 ## 建立 Dataset 1. 複製 image, label ( 順便校準 bbox ) 2. 將 label 做標準化 3. 儲存成以下格式 :::warning > - datasets > - images > - train > - 00001.png > - ... > - labels > - train > - 00001.txt > - ... > ::: <br> ## 訓練 Model ```python= # 訓練 !python train.py --epochs 100 --batch-size 64 --data drone.yaml --cfg yolov5s.yaml --weights '' --workers 0 ``` - `--epochs EPOCHS` : 訓練次數 - `--batch-size BATCH_SIZE` - `--data DATA資訊檔(.yaml)` - `--cfg Model檔(.yaml)` - `--workers WORKER數量` ```python= # 查看所有訓練參數 !python train.py --help ``` 訓練出的結果存放在 **./runs/train/expN/** <br> ## Data Yaml 檔設定 檔案要存放在 **./data/** ```yaml= # 路徑 path: ../datasets/drone train: images/train val: images/train test: # 物件類別設定 nc: 4 names: ['car','hov','person','motorcycle'] ``` <br> ## Validate Model ```python= !python val.py --weights runs/train/exp4/weights/best.pt --data drone.yaml --half --workers 0 ``` 結果存放在 **./runs/val/expN/** <br> ## 用 Model 進行偵測 ```python= !python detect.py --weights runs/train/exp4/weights/best.pt --conf 0.25 --source ../public ``` 結果存放在 **./runs/detect/expN/**