(like Google Open-Images-v7),
Datasets, available in fuftyone detasets
- Edit config.py
- Download images from open-images-v7:
python download_dataset.py
python export_dataset_yolo_v5.py
dataset.yaml example structure:
train: /home/ubuntu/train/dataset_yolo/images/train
val: /home/ubuntu/train/dataset_yolo/images/validation
test: /home/ubuntu/train/dataset_yolo/images/test
nc: 2 # Number of classes (change to match your number of classes)
names: /home/ubuntu/train/dataset_yolo/classes.txt
git clone https://github.com/ultralytics/yolov5.git
python train.py --img 416 --batch 32 --epochs 1 --data /home/ubuntu/train/dataset_yolo/dataset.yaml --weights yolov5s.pt
rsync -avz frod:/home/ubuntu/train/yolov5/runs/train/exp9/weights/best.pt ~/Desktop
5. Convert results with Luxonis Model Converter
git clone git@github.com:luxonis/depthai-experiments.git
cp ~/Downloads/model/* depthai-experiments/gen2-yolo/device-decoding/model/
python main_api.py \
--config model/1000best/best.json \
--model model/1000best/best_openvino_2022.1_6shave.blob