Result of `model.predict` has type `ClassificationModel`, no keypoints.

Tested on several projects:

• Glue tube keypoints: https://app.roboflow.com/model-examples/glue-tube-keypoints
• Football pitch recognition below.

pip install supervision roboflow

version = rf.workspace("roboflow-jvuqo").project("football-field-detection-f07vi").version(8)
version.download("yolov8")
kp_model = version.model
assert kp_model is not None

first_image_path = os.listdir(f"{dataset.location}/train/images")[0]
first_image_path = f"{dataset.location}/train/images/{first_image_path}"

result = kp_model.predict(first_image_path).json()

# Fails
keypoints = sv.KeyPoints.from_inference(result)

Riproduci il report con il comando pip install fornito e l’esempio del modello football-field, quindi ispeziona il risultato di kp_model.predict(first_image_path).json(). Confronta quell’output con l’input previsto da sv.KeyPoints.from_inference(result); il lavoro è completato quando il risultato dell’inferenza dei keypoint è rappresentato in modo tale che questa chiamata non fallisca più.