YOLO v9 - Object Detection Gets a New Upgrade

You Only Look Once (YOLO) is one of the most well-known model architectures to have dominated the computer vision space. Its object detection algorithm is renowned for fast image processing results and improved accuracy. The YOLO algorithm strives to predict an object’s class and the bounding box that pinpoints its location on the input image.

Several iterations of YOLO have been released since Joseph Redmon first introduced it in 2015; the most current was created by AI platform Ultralytics, who also made versions YOLO v3 and YOLO v5. 

YOLO Details

YOLO	You Only Look Once
Latest Version	V9
Release Date	February 21, 2024
Developed by	Joseph Redmon and Santosh Divvala
Source Codes	https://github.com/WongKinYiu/yolov9

All About YOLO V9

YOLO v9 emerges as a cutting-edge model, boasting innovative features that will play an important role in the further development of object detection, image segmentation, and classification. The new top-tier features allow faster, sharper, and more versatile actions. 

The latest research paper proposed the use of Programmable Gradient Information (PGI) to tackle the information bottleneck and the challenge of adapting deep supervision to lightweight architectures of neural networks. 

The team has also designed the Generalized Efficient Layer Aggregation Network (GELAN), a handy and practically effective neural network. The network has strong and stable performance at different computational blocks and depth settings regarding object detection. It can indeed be widely expanded into a model suitable for various inference devices. 

For the above two issues, the introduction of PGI allows both lightweight as well as deep models to achieve significant improvements in accuracy. The YOLO v9, designed by combining PGI and GELAN, has shown strong competitiveness. Its well-thought design allows the deep model to reduce the number of parameters by 49% and the amount of calculations by 43% compared with YOLO v8. However, it still has a 0.6% Average Precision improvement on the MS COCO dataset.

The latest Yolo-v model beats RT-DETR (Realtime Detection Transformer) and YOLO MS in accuracy and efficiency. It uses conventional convolution for better parameter utilization, setting new standards in lightweight model performance.

Here’s the source code for Yolo v9.