sv.Detections.from_transformers expects different dict-properties about supervision

[KevinHubert-Dev]

Found the right code snippet to use in the "inference" chapter of Huggingface
It is supposed to be similiar to this one

with torch.no_grad():
    inputs = image_processor(images=image, return_tensors="pt")
    outputs = model(**inputs)
    target_sizes = torch.tensor([image.size[::-1]])
    results = image_processor.post_process_object_detection(outputs, threshold=0.5, target_sizes=target_sizes)[0]

No changes needed. Just be aware that you need to use the plain model-call approach because the pipeline returns as different structured format