目標檢測後處理：NMS實現

Pytorch、Tensorflow中均內置了NMS能力的函數。這裏給出兩種基於tensorflow/keras的實現，第一種來自EfficientDet，GPU版本支持正框和斜框。

from tensorflow import keras
import tensorflow as tf

def filter_detections(
        boxes,
        classification,
        alphas=None,
        ratios=None,
        class_specific_filter=True,
        nms=True,
        score_threshold=0.01,
        max_detections=100,
        nms_threshold=0.5,
        detect_quadrangle=False,
):
    """
    Filter detections using the boxes and classification values.
    Args
        boxes: Tensor of shape (num_boxes, 4) containing the boxes in (x1, y1, x2, y2) format.
        classification: Tensor of shape (num_boxes, num_classes) containing the classification scores.
        other: List of tensors of shape (num_boxes, ...) to filter along with the boxes and classification scores.
        class_specific_filter: Whether to perform filtering per class, or take the best scoring class and filter those.
        nms: Flag to enable/disable non maximum suppression.
        score_threshold: Threshold used to prefilter the boxes with.
        max_detections: Maximum number of detections to keep.
        nms_threshold: Threshold for the IoU value to determine when a box should be suppressed.
    Returns
        A list of [boxes, scores, labels, other[0], other[1], ...].
        boxes is shaped (max_detections, 4) and contains the (x1, y1, x2, y2) of the non-suppressed boxes.
        scores is shaped (max_detections,) and contains the scores of the predicted class.
        labels is shaped (max_detections,) and contains the predicted label.
        other[i] is shaped (max_detections, ...) and contains the filtered other[i] data.
        In case there are less than max_detections detections, the tensors are padded with -1's.
    """

    def _filter_detections(scores_, labels_):
        # threshold based on score
        # (num_score_keeps, 1)
        indices_ = tf.where(keras.backend.greater(scores_, score_threshold))

        if nms:
            # (num_score_keeps, 4)
            filtered_boxes = tf.gather_nd(boxes, indices_)
            # In [4]: scores = np.array([0.1, 0.5, 0.4, 0.2, 0.7, 0.2])
            # In [5]: tf.greater(scores, 0.4)
            # Out[5]: <tf.Tensor: id=2, shape=(6,), dtype=bool, numpy=array([False,  True, False, False,  True, False])>
            # In [6]: tf.where(tf.greater(scores, 0.4))
            # Out[6]:
            # <tf.Tensor: id=7, shape=(2, 1), dtype=int64, numpy=
            # array([[1],
            #        [4]])>
            #
            # In [7]: tf.gather(scores, tf.where(tf.greater(scores, 0.4)))
            # Out[7]:
            # <tf.Tensor: id=15, shape=(2, 1), dtype=float64, numpy=
            # array([[0.5],
            #        [0.7]])>
            filtered_scores = keras.backend.gather(scores_, indices_)[:, 0]

            # perform NMS
            # filtered_boxes = tf.concat([filtered_boxes[..., 1:2], filtered_boxes[..., 0:1],
            #                             filtered_boxes[..., 3:4], filtered_boxes[..., 2:3]], axis=-1)
            nms_indices = tf.image.non_max_suppression(filtered_boxes, filtered_scores, max_output_size=max_detections,
                                                       iou_threshold=nms_threshold)

            # filter indices based on NMS
            # (num_score_nms_keeps, 1)
            indices_ = keras.backend.gather(indices_, nms_indices)

        # add indices to list of all indices
        # (num_score_nms_keeps, )
        labels_ = tf.gather_nd(labels_, indices_)
        # (num_score_nms_keeps, 2)
        indices_ = keras.backend.stack([indices_[:, 0], labels_], axis=1)

        return indices_

    if class_specific_filter:
        all_indices = []
        # perform per class filtering
        for c in range(int(classification.shape[1])):
            scores = classification[:, c]
            labels = c * tf.ones((keras.backend.shape(scores)[0],), dtype='int64')
            all_indices.append(_filter_detections(scores, labels))

        # concatenate indices to single tensor
        # (concatenated_num_score_nms_keeps, 2)
        indices = keras.backend.concatenate(all_indices, axis=0)
    else:
        scores = keras.backend.max(classification, axis=1)
        labels = keras.backend.argmax(classification, axis=1)
        indices = _filter_detections(scores, labels)

    # select top k
    scores = tf.gather_nd(classification, indices)
    labels = indices[:, 1]
    scores, top_indices = tf.nn.top_k(scores, k=keras.backend.minimum(max_detections, keras.backend.shape(scores)[0]))

    # filter input using the final set of indices
    indices = keras.backend.gather(indices[:, 0], top_indices)
    boxes = keras.backend.gather(boxes, indices)
    labels = keras.backend.gather(labels, top_indices)

    # zero pad the outputs
    pad_size = keras.backend.maximum(0, max_detections - keras.backend.shape(scores)[0])
    boxes = tf.pad(boxes, [[0, pad_size], [0, 0]], constant_values=-1)
    scores = tf.pad(scores, [[0, pad_size]], constant_values=-1)
    labels = tf.pad(labels, [[0, pad_size]], constant_values=-1)
    labels = keras.backend.cast(labels, 'int32')

    # set shapes, since we know what they are
    boxes.set_shape([max_detections, 4])
    scores.set_shape([max_detections])
    labels.set_shape([max_detections])

    if detect_quadrangle:
        alphas = keras.backend.gather(alphas, indices)
        ratios = keras.backend.gather(ratios, indices)
        alphas = tf.pad(alphas, [[0, pad_size], [0, 0]], constant_values=-1)
        ratios = tf.pad(ratios, [[0, pad_size]], constant_values=-1)
        alphas.set_shape([max_detections, 4])
        ratios.set_shape([max_detections])
        return [boxes, scores, alphas, ratios, labels]
    else:
        return [boxes, scores, labels]

另一個CPU版本，支持正框

https://blog.csdn.net/m0_37605642/article/details/98358864