W3cubDocs

/TensorFlow 2.4

tf.image.non_max_suppression_padded

Greedily selects a subset of bounding boxes in descending order of score.

Performs algorithmically equivalent operation to tf.image.non_max_suppression, with the addition of an optional parameter which zero-pads the output to be of size max_output_size. The output of this operation is a tuple containing the set of integers indexing into the input collection of bounding boxes representing the selected boxes and the number of valid indices in the index set. The bounding box coordinates corresponding to the selected indices can then be obtained using the tf.slice and tf.gather operations. For example: ```python selected_indices_padded, num_valid = tf.image.non_max_suppression_padded( boxes, scores, max_output_size, iou_threshold, score_threshold, pad_to_max_output_size=True) selected_indices = tf.slice( selected_indices_padded, tf.constant([0]), num_valid) selected_boxes = tf.gather(boxes, selected_indices)

Args
boxes a tensor of rank 2 or higher with a shape of [..., num_boxes, 4]. Dimensions except the last two are batch dimensions.
scores a tensor of rank 1 or higher with a shape of [..., num_boxes].
max_output_size a scalar integer Tensor representing the maximum number of boxes to be selected by non max suppression.
iou_threshold a float representing the threshold for deciding whether boxes overlap too much with respect to IoU (intersection over union).
score_threshold a float representing the threshold for box scores. Boxes with a score that is not larger than this threshold will be suppressed.
pad_to_max_output_size whether to pad the output idx to max_output_size. Must be set to True when the input is a batch of images.
name name of operation.
sorted_input a boolean indicating whether the input boxes and scores are sorted in descending order by the score.
canonicalized_coordinates if box coordinates are given as [y_min, x_min, y_max, x_max], setting to True eliminate redundant computation to canonicalize box coordinates.
tile_size an integer representing the number of boxes in a tile, i.e., the maximum number of boxes per image that can be used to suppress other boxes in parallel; larger tile_size means larger parallelism and potentially more redundant work.
Returns
idx: a tensor with a shape of [..., num_boxes] representing the indices selected by non-max suppression. The leading dimensions are the batch dimensions of the input boxes. All numbers are within [0, num_boxes). For each image (i.e., idx[i]), only the first num_valid[i] indices (i.e., idx[i][:num_valid[i]]) are valid. num_valid: a tensor of rank 0 or higher with a shape of [...] representing the number of valid indices in idx. Its dimensions are the batch dimensions of the input boxes.
Raises ValueError: When set pad_to_max_output_size to False for batched input.

© 2020 The TensorFlow Authors. All rights reserved.
Licensed under the Creative Commons Attribution License 3.0.
Code samples licensed under the Apache 2.0 License.
https://www.tensorflow.org/versions/r2.4/api_docs/python/tf/image/non_max_suppression_padded