What is YOLO

YOLO (You Only Look Once) is a state-of-the-art, real-time object detection system. It predicts bounding boxes and class probabilities for objects in an image with high accuracy and speed. YOLO models, including the latest YOLOv9, are known for their efficiency in detecting objects in a single forward pass, making them highly suitable for real-time applications.

YOLOv9 introduces improvements in both architecture and loss functions to enhance prediction accuracy and inference speed.

Forward Process

graph LR
subgraph YOLOv9
    Auxiliary
    AP["Auxiliary Prediction"]
end
BackBone-->FPN;
FPN-->PAN;
PAN-->MP["Main Prediction"];
BackBone-->Auxiliary;
Auxiliary-->AP;

• BackBone: Extracts features from the input image.
• FPN (Feature Pyramid Network): Aggregates features at different scales.
• PAN (Region Proposal Network): Proposes regions of interest.
• Main Prediction: The primary detection output.
• Auxiliary Prediction: Additional predictions to assist the main prediction.

Loss Function

flowchart LR
    gtb-->cls
    gtb["Ground Truth"]-->iou
    pdm-.->cls["Max Class"]
    pdm["Main Prediction"]-.->iou["Closest IoU"]
    pdm-.->anc["box in anchor"]
    cls-->gt
    iou-->gt["Matched GT Box"]
    anc-.->gt

    gt-->Liou["IoU Loss"]
    pdm-->Liou
    pdm-->Lbce
    gt-->Lbce["BCE Loss"]
    gt-->Ldfl["DFL Loss"]
    pdm-->Ldfl

    Lbce-->ML
    Liou-->ML
    Ldfl-->ML["Total Loss"]

• Ground Truth: The actual labels and bounding boxes in the dataset.
• Main Prediction: The model's predicted bounding boxes and class scores.
• IoU (Intersection over Union): Measures overlap between predicted and ground truth boxes.
• BCE (Binary Cross-Entropy) Loss: Used for class prediction.
• DFL (Distribution Focal Loss): Improves precision of bounding box regression.