Custom arguments for Yolo-v8 model

[HT0710]

The original Yolo-v8 model offer some arguments like filter classes or half,... which is unable to access through AutoDetectionModel.from_pretrained

Anyone have any ideas?

[GuillaumeBruand]

@HT0710 Custom arguments from config file are used at inference time in yolov8.

You will need a little customization for this until a PR is done, here is my implementation:

from sahi.models.yolov8 import Yolov8DetectionModel
class Yolov8CustomDetectionModel(Yolov8DetectionModel):
    def perform_inference(self, image):
        """
        Prediction is performed using self.model and the prediction result is set to self._original_predictions.
        Args:
            image: np.ndarray
                A numpy array that contains the image to be predicted. 3 channel image should be in RGB order.
        """

        # Confirm model is loaded
        if self.model is None:
            raise ValueError("Model is not loaded, load it by calling .load_model()")
        prediction_result = self.model(image[:, :, ::-1],
                                       cfg=self.config_path,
                                       verbose=False,
                                       conf=self.confidence_threshold)  # YOLOv8 expects numpy arrays to have BGR
        prediction_result = [
            result.boxes.data for result in prediction_result
        ]

        self._original_predictions = prediction_result

Notice that I also added the "conf" argument for the inner yolo model to use the provided threshold.
You can use the traditional yolov8 yaml config file and pass it using the config file argument:

detection_model = Yolov8CustomDetectionModel(
        model_path="my_model_path_here",
        confidence_threshold=0.2,
        device="cpu",  # or 'cuda:0'
        image_size=256,
        config_path="yolo_config_file.yaml"
    )

[HT0710]

Thanks!!

[NawoyaSarah]

Hi, can i use this in instance segmentation where am working with masks?

[filipemendesdev]

I have the same question