Welcome to the YOLO-NAS Repository!
Object Detection
Our architecture is designed to deliver unparalleled accuracy-speed performance, pushing the boundaries of what's possible in object detection. In this repository, we provide instructions for training your own model using our cutting-edge architecture.
To begin your training journey, first install the super-gradients library version 3.1.1 with the following command:
!pip install -q super-gradients==3.1.1
!pip install onemetric
!pip install -q supervision
After installation is complete, don't forget to restart the runtime by navigating to Runtime -> Restart runtime and confirming with a simple click of "Yes".
Next, clone this repository with the following command:
!git clone https://github.com/VYRION-Ai/Yolo-Nas.git
You'll then want to add your dataset to the mix. To do this, use the following code snippet with Google Colab:
from google.colab import drive
drive.mount('/content/drive')
%cd /content/
%cp /content/drive/MyDrive/yolo_dataset.zip /content/
!unzip /content/yolo_dataset.zip
!rm /content/yolo_dataset.zip
And finally, it's time to start training! Navigate to the YOLO-NAS directory and run the following command to begin your project:
%cd /content/Yolo-Nas
!python train.py --project "Dataset" --data /content/Dataset/data.yaml --location '/content/Dataset' --model-arch yolo_nas_s --batch-size 16 --max-epochs 25 --checkpoint-dir /content/checkpoints
Evaluate trained model
%cd /content/Yolo-Nas
!python valid.py --data /content/Dataset/data.yaml --location '/content/Dataset' --project "Dataset" --weights /content/checkpoints/Dataset/ckpt_best.pth
We welcome contributions from the community! If you encounter any issues or have any suggestions for improving the YOLO-NAS architecture, please feel free to open an issue or submit a pull request.
This repository is licensed under the MIT license. See LICENSE for more information.