tfm.vision.factory.build_retinanet

Builds a RetinaNet model.

tfm.vision.factory.build_retinanet(
    input_specs: tf_keras.layers.InputSpec,
    model_config: tfm.vision.configs.retinanet.RetinaNet,
    l2_regularizer: Optional[tf_keras.regularizers.Regularizer] = None,
    backbone: Optional[tf_keras.Model] = None,
    decoder: Optional[tf_keras.Model] = None,
    num_anchors_per_location: (int | dict[str, int] | None) = None,
    anchor_boxes: (Mapping[str, tf.Tensor] | None) = None
) -> tf_keras.Model

input_specs The InputSpec of the input image tensor to the model.
model_config The RetinaNet model configuration to build from.
l2_regularizer Optional l2 regularizer to use for building the backbone, decorder, and head.
backbone Optional instance of the backbone model.
decoder Optional instance of the decoder model.
num_anchors_per_location Optional number of anchors per pixel location for building the RetinaNetHead. If an int, the same number is used for all levels. If a dict, it specifies the number at each level. If none, it uses len(aspect_ratios) * num_scales from the anchor config by default.
anchor_boxes Optional fixed multilevel anchor boxes for inference.

RetinaNet model.