Retinanet implementing COCOMeanAveragePrecision metric (kerasCV)

I implemented the Object Detection with RetinaNet Code by https://keras.io/examples/vision/retinanet/.

Now i want to import the Metric MeanAveragePrecision. I try to use the keras_cv.metrics.COCOMeanAveragePrecision as metric but i get for mAP and REcall 0. Can someone tell me what i do wrong, or is there another way to get the Map or IoU metric?

I put the metric call in my model compile

resnet50_backbone = get_backbone()
loss_fn = RetinaNetLoss(num_classes)
model = RetinaNet(num_classes, resnet50_backbone)


optimizer = tf.optimizers.SGD(learning_rate=learning_rate_fn, momentum=0.9)
model.compile(loss=loss_fn, optimizer=optimizer, metrics=[keras_cv.metrics.COCOMeanAveragePrecision(
    bounding_box_format="xyxy",
    max_detections=100,
    class_ids=range(80),
    area_range=(0, 64**2),
    name="Mean Average Precision",
)])

the resnet50_backbone, loss_fn, and model ist from the keras implementation i linked at the top