YOLO-NAS is a new State of the Art, foundation model for object detection inspired by YOLOv6 and YOLOv8.
Here's why you've got to give it a try:
🧱 New Quantization-Friendly Block: Improving on previous models, YOLO-NAS features a novel basic block that's tailor-made for quantization.
🚀 Advanced Training Scheme: YOLO-NAS undergoes pre-training on the Object365 dataset, leverages pseudo-labeled data, and benefits from knowledge distillation using a pre-trained teacher model.
🎯 Post-Training Quantization (PTQ): The network is converted to INT8 after training, making it even more efficient.
🧬 AutoNac Optimized: Three final networks emerge from applying AutoNac on the architecture space, all while using the equivalent GPU time of training just 5 networks.
💾 Pre-Trained on Top Datasets: YOLO-NAS comes pre-trained on COCO, Objects365, and Roboflow 100, setting you up for success in downstream object detection tasks.
Deci's open-source, PyTorch-based computer vision library, SuperGradients, makes YOLO-NAS easy to train, and has advanced techniques like Distributed Data Parallel, Exponential Moving Average, Automatic Mixed Precision, and Quantization Aware Training.
🌟 Head over to GitHub, star SuperGradients, and explore the starter notebook to see YOLO-NAS in action!
# Clone repo
git clone "https://github.com/TunaUlusoy/YOLO-NAS.git"
cd YOLO-NAS
# Create environmet
conda env create -f yolonas.yml
conda activate yolonas
Technical blog: https://deci.ai/blog/YOLO-NAS-object-detection-foundation-model
GitHub repo: https://github.com/Deci-AI/super-gradients/blob/master/YOLONAS.md
Starter Notebook: https://colab.research.google.com/drive/1q0RmeVRzLwRXW-h9dPFSOchwJkThUy6d