Segmentation Support

[donalSE]

Hi,

Just wondering is segmentation fully operational with SAHI at the moment? I have a custom segmentation model, however when I run it, predictions always return with 'mask=None'.

I've read through all segmentation related topics on the Discussions board and it seems as if this problem isn't new. Also quite sure everything is setup to integrate masks should they be detected. Interested to hear any thoughts or suggestions.

Code:
``
class SahiDetection:
def init(self, model_path: str = 'seg-5epoch.pt', config_path: str = 'data.yaml'):
self.device = 'cuda' if torch.cuda.is_available() else 'cpu'
print("Using Device: ", self.device)
self.model = self.load_model(model_path, config_path)

def load_model(self, model_path, config_path):
    model = AutoDetectionModel.from_pretrained(
        model_type='yolov8',
        model_path=model_path,
        config_path=config_path,
    )
    return model

def predict(self, frame: np.ndarray):
    height, width, _ = frame.shape
    results = get_sliced_prediction(
        frame,
        self.model,
        slice_height=270,
        slice_width=480,
        overlap_height_ratio=0.2,
        overlap_width_ratio=0.2
    )
    print(results.object_prediction_list)
    return results

Subsequent Print statement:

bbox: BoundingBox: <(181.3393096923828, 648, 210.01722717285156, 698.9091377258301), w: 28.67791748046875, h: 50.90913772583008>,
mask: None,
score: PredictionScore: <value: 0.6274577379226685>,
category: Category: <id: 4, name: hidden>>]
``

[fcakyon]

Segmentation mask support for yolov8, mmdetection, detectron2, torchvision models is live with the latest release! @donalSE https://github.com/obss/sahi/discussions/1051

There is also 5x memory and speed improvement when working with masks!

Check the demo notebook on how it works: https://github.com/obss/sahi/blob/main/demo/inference_for_yolov8.ipynb