Build Model¶
In YOLOv7, predictions are Anchor-based. In YOLOv9, predictions are Vector-based. A converter transforms bounding boxes to the appropriate format.
flowchart LR
Input-->Model;
Model--Class-->NMS;
Model--Anc/Vec-->Converter;
Converter--Box-->NMS;
NMS-->Output;Load Model¶
Use create_model to automatically create the YOLO model and load weights.
| Argument | Type | Description |
|---|---|---|
model |
ModelConfig |
The model configuration |
class_num |
int |
Number of dataset classes, used for the prediction head |
weight_path |
Path \| bool |
False = no weights; True/None = default weights; Path = load from path |
model = create_model(cfg.model, class_num=cfg.dataset.class_num, weight_path=cfg.weight)
model = model.to(device)
Deploy Model¶
Removes the auxiliary branch for fast inference. Loads/compiles to ONNX or TensorRT if configured.
Autoload Converter¶
Autoloads the converter based on model type (v7 → Anc2Box, v9 → Vec2Box).
| Argument | Type | Description |
|---|---|---|
| Model Name | str |
Selects Vec2Box or Anc2Box |
| Anchor Config | — | Anchor configuration for generating the anchor grid |
model, image_size |
— | Used for auto-detecting the anchor grid |