This is a basic setup to train a yolo model set up a virtual environment python3 -m venv <folder>
source bin/activate
getting data sets from "Open Images" using "OIDv4_ToolKit"
OIDv4_ToolKit is a tool that downloads data sets from open images based on classes
setting up a seperate virtual environment for "OIDv4" is recommended
git clone https://github.com/EscVM/OIDv4_ToolKit.git
python3 main.py downloader --classes Person --type_csv train --limit 1000
python3 main.py downloader --classes Person --type_csv test --limit 300
make sure to download the csv annotations with the dataset
.
├── csv_folder
└── Dataset
├── test
│ ├── images
│ └── labels
└── train
├── images
└── labels
setting up the training environment
copy the downloaded csv annotations to the csv_folder
copy both downloaded training and testing images to train/images and test/images
copy both downloaded labels to /train/labels and /test/labels
run convert.py to covert the labels into YOLO format
download dependencys
pip install -r requirements.txt
using the model in real time video caption