tensorflow利用矩阵计算实现一维数据的NMS用于TimeSerialDectection

最新推荐文章于 2020-03-26 00:06:52 发布

AR_SAMUEL

最新推荐文章于 2020-03-26 00:06:52 发布

阅读量319

点赞数

分类专栏：深度学习文章标签： tensorflow NMS 目标检测

本文链接：https://blog.csdn.net/AR_SAMUEL/article/details/100775129

版权

本文详细介绍了如何使用TensorFlow库在一维数据上应用非极大值抑制（NMS）技术，以进行高效的目标检测，特别适用于时序数据的检测任务。

摘要由CSDN通过智能技术生成

# 行列式相乘可以有循环遍历的效果
import tensorflow as tf
import numpy as np
def compute_overlaps(boxes):
    """Computes IoU overlaps between two sets of boxes.
    boxes1, boxes2: [N, (x1, x2)].

    For better performance, pass the largest set first and the smaller second.
    """
    b1_x1, b1_x2 = \
        tf.expand_dims(boxes[:, 0],axis=-2), \
        tf.expand_dims(boxes[:, 1],axis=-2),

    b2_x1, b2_x2 = \
        tf.expand_dims(boxes[:, 0],axis=-1),\
        tf.expand_dims(boxes[:, 1],axis=-1),


    # get the corrdinates of the intersection
    inter_rect_x1 = tf.maximum(b1_x1, b2_x1)

    inter_rect_x2 = tf.minimum(b1_x2, b2_x2)

    # Intersection area
    inter_area = tf.abs(inter_rect_x2 - inter_rect_x1 + 1)

    # Union Area
    b1_area = tf.abs((b1_x2 - b1_x1 + 1))
    b2_area = tf.abs((b2_x2 - b2_x1 + 1))

    iou = inter_area / (b1_area + b2_a