SSD class

dlf.models.ssd.SSD(
    num_classes,
    arch="vgg_ssd_300",
    input_shape=(300, 300, 3),
    weight_decay=0.0005,
    optimizer=None,
    loss=None,
    model_weights=None,
    summary=False,
    ratios=None,
    scales=None,
    class_score_threshold=0.6,
    nms_threshold=0.45,
    max_boxes=200,
    **kwargs
)

A implementation for an Single Shot Detector (SSD)

Aliases

• ssd
• SSD

Architectures implemented

• vgg_ssd_300: Original Paper implmentation of SSD 300 with VGG as backend
• vgg_ssd_512: Original Paper implmentation of SSD 512 with VGG as backend

Args

• num_classes:
• arch: str. SSD architecture. Defaults to 'vgg_ssd_300'.
• input_shape: tuple(int, int , int). Input shape of this network. Defaults to (300, 300, 3).
• weight_decay: float. Weight decay. Defaults to 5e-4.
• optimizer: list of dict, optional. Name of optimizer used for training.
• loss: list of dict, optional. Not Implemented Yet! List of loss objects to build for this model. Defaults to None.
• model_weights: str, optional. Path to the pretrained model weights. Defaults to None.
• summary: bool, optional. If true a summary of the model will be printed to stdout. Defaults to False.
• ratios: list of list of float, optional. List of anchor box ratios for SSD. Defaults to None.
• scales: list of float. List of scalings for anchor boxes. Defaults to None.
• class_score_threshold: float, optional. Class confidence threshold . Defaults to 0.6.
• nms_threshold: float. Min IoU for non-maximum suppression. Defaults to 0.45.
• max_boxes: int, optional. Max number of Boxes. Defaults to 200.

Returns

A Keras model instance.

YAML Configuration

model:
    ssd:
        num_classes: &num_classes 7
        input_shape:
        - 512
        - 512
        - 3
        summary: True
        optimizer:
        - Adam:
            learning_rate: 0.001
        class_score_threshold: 0.6
        nms_threshold: 0.45
        max_boxes: 200
        arch: vgg_ssd_512

References

• Single Shot Detector https://arxiv.org/abs/1512.02325

---