TRAINING.md

Training the YOLOv5 Network

Downloading the Pre-trained Network

Download YOLOv5n6 and store it in weights.

Preparing the Data

Download the labeled data and unpack it. Edit the first row of the file merged.yaml replacing the placeholder path that is given there by the absolute path to the directory that contains the file merged.yaml

Training

From the main directory of this repository, run (replace /path/to/ and adapt --batch if necessary):

yolov5 train --weights weights/yolov5n6.pt --data /path/to/merged.yaml --img 1920 --batch 8 --epochs 400

The results will be written to runs/train/exp*. Further instructions assume that the network was written to runs/train/exp.

Detection

Run:

yolov5 detect --weights runs/train/exp/weights/best.pt --data /path/to/merged.yaml --img 1088,1920 --line-thickness 2 --hide-labels --conf 0.3 --source /path/to/image_dir_or_video

The results will be written to runs/detect/exp*.

Validation

To get some statistics about the training results, run:

yolov5 val --weights runs/train/exp/weights/best.pt --data /path/to/merged.yaml --img 1920 --task test

The results will be written to runs/val/exp*.

Committing a Trained Model

To finally commit a trained model, copy runs/train/exp/weights/best.pt to weights in the main directory of this repository.