The Object detection model customization Samples model does not support MediaPipe. Models are expected to have exactly four outputs, but only two were found.

[liangbuwen]

The Object detection model customization Samples model does not support MediaPipe. Models are expected to have exactly 4 outputs, but only 2 were found.

I ran the code in the Object detection model customization guide to train a custom model

spec = object_detector.SupportedModels.MOBILENET_MULTI_AVG
hparams = object_detector.HParams(export_dir='exported_model')
options = object_detector.ObjectDetectorOptions(
    supported_model=spec,
    hparams=hparams
)

However, my custom model is not supported in the MediaPipe example.
Because the SupportedModels supported by the object detector solution are all trained with only 2 outputs:
detection_boxes and detection_scores,
which is inconsistent with the output format required of the model . ( outputs format required detection_boxes,detection_classes,detection_scores,num_boxes 4 outputs
How can I retrain a MobileNetv2 model to have four outputs that meet the requirements?