How do I know or decide the field-names on Checkpoint constructor?

Hey there,

In interactive_eager_few_shot_od_training_colab I see the code piece:

fake_box_predictor = tf.compat.v2.train.Checkpoint(
    _base_tower_layers_for_heads=detection_model._box_predictor._base_tower_layers_for_heads,
    # _prediction_heads=detection_model._box_predictor._prediction_heads,
    #    (i.e., the classification head that we *will not* restore)
    _box_prediction_head=detection_model._box_predictor._box_prediction_head,
    )

fake_model = tf.compat.v2.train.Checkpoint(
          _feature_extractor=detection_model._feature_extractor,
          _box_predictor=fake_box_predictor)

ckpt = tf.compat.v2.train.Checkpoint(model=fake_model)

I have checked the ssd_resnet50_v1_fpn_640x640_coco17_tpu-8.config, I do not see the associated namings, like _box_prediction_head , _base_tower_layers_for_heads etc.

Maybe the question is fool, but I was completely overwhelmed and baffled by this lab.

Needing help, thanks.

Hi there,

In this weeks assignment you will see references to the github repositories where relevant functions are implemented and these names are extracted from that code refering to parts of the implementation.

1 Like