tfm.vision.anchor.RpnAnchorLabeler

Labeler for Region Proposal Network.

Inherits From: AnchorLabeler

tfm.vision.anchor.RpnAnchorLabeler(
    match_threshold=0.7,
    unmatched_threshold=0.3,
    rpn_batch_size_per_im=256,
    rpn_fg_fraction=0.5
)

match_threshold a float number between 0 and 1 representing the lower-bound threshold to assign positive labels for anchors. An anchor with a score over the threshold is labeled positive.
unmatched_threshold a float number between 0 and 1 representing the upper-bound threshold to assign negative labels for anchors. An anchor with a score below the threshold is labeled negative.
box_coder_weights Optional list of 4 positive floats to scale y, x, h, and w when encoding box coordinates. If set to None, does not perform scaling. For Faster RCNN, the open-source implementation recommends using [10.0, 10.0, 5.0, 5.0].

Methods

label_anchors

View source

label_anchors(
    anchor_boxes: Dict[str, tf.Tensor],
    gt_boxes: tf.Tensor,
    gt_labels: tf.Tensor
) -> Tuple[Dict[str, tf.Tensor], Dict[str, tf.Tensor]]

Labels anchors with ground truth inputs.

Args
anchor_boxes An ordered dictionary with keys [min_level, min_level+1, ..., max_level]. The values are tensor with shape [height_l, width_l, num_anchors_per_location * 4]. The height_l and width_l represent the dimension of the feature pyramid at l-th level. For each anchor box, the tensor stores [y0, x0, y1, x1] for the four corners.
gt_boxes A float tensor with shape [N, 4] representing ground-truth boxes. For each row, it stores [y0, x0, y1, x1] for four corners of a box.
gt_labels A integer tensor with shape [N, 1] representing ground-truth classes.

Returns
score_targets_dict An ordered dictionary with keys [min_level, min_level+1, ..., max_level]. The values are tensor with shape [height_l, width_l, num_anchors_per_location]. The height_l and width_l represent the dimension of class logits at l-th level.
box_targets_dict An ordered dictionary with keys [min_level, min_level+1, ..., max_level]. The values are tensor with shape [height_l, width_l, num_anchors_per_location * 4]. The height_l and width_l represent the dimension of bounding box regression output at l-th level.