1. 基本思想 

NMS（Non-Maximum Suppression）算法本质是搜索局部极大值，抑制非极大值元素。NMS就是需要根据score矩阵和region的坐标信息，从中找到置信度比较高的bounding box。当两个box空间位置非常接近，就以score更高的那个作为基准，看IOU即重合度如何，如果与其重合度超过阈值，就抑制score更小的box，因为没有必要输出两个接近的box，只保留score大的就可以了。

2. 计算流程

1. 所有检测框按置信度从高到低排序

2. 取当前置信度最高的框，然后删除和这个框的iou高于阈值的框

3. 重复第2步直到所有框处理完。

3. 存在问题

• 顺序处理的模式，计算IoU拖累了运算效率。
• 剔除机制太严格，依据NMS阈值暴力剔除。
• 阈值是经验选取的。
• 评判标准是IoU，即只考虑两个框的重叠面积，这对描述box重叠关系或许不够全面。

4. 代码实现

import numpy as npclass multiclass_nms():"""bboxes: [N，M，4]，表示将预测M个边界框的预测位置， N是batch size。每个边界框有四个坐标值，布局为[xmin，ymin，xmax，ymax]。scores: [N，C，M]，表示预测的置信度。 N 是 batch size，C是种类数目，M是边界框bounding box的数量。对于每个类别，存在对应于M个边界框的总M个分数。score_thresh: 过滤掉低置信度分数的边界框的阈值。如果没有提供，请考虑所有边界框。pos_nms_topk: NMS步骤后每个图像要保留的总bbox数。nms_thresh: 在NMS中用于剔除检测框IOU的阈值。"""def __init__(self, score_thresh=0.1, nms_thresh=0.5, pos_nms_topk=120):self.score_thresh = score_threshself.nms_thresh = nms_threshself.pos_nms_topk = pos_nms_topkdef multiclass_nms_(self, boxes, scores):batch_size = boxes.shape[0]class_num = scores.shape[1]rets = []for i in range(batch_size):bboxes_i = boxes[i]scores_i = scores[i]ret = []# 遍历所有类别，进行单分类NMSfor c in range(class_num):scores_i_c = scores_i[c]keep_inds = self.nms(bboxes_i, scores_i_c)if len(keep_inds) < 1:continuekeep_bboxes = bboxes_i[keep_inds]keep_scores = scores_i_c[keep_inds]keep_results = np.zeros([keep_scores.shape[0], 6])keep_results[:, 0] = ckeep_results[:, 1] = keep_scores[:]keep_results[:, 2:6] = keep_bboxes[:, :]ret.append(keep_results)if len(ret) < 1:rets.append(ret)continueret_i = np.concatenate(ret, axis=0)scores_i = ret_i[:, 1]if len(scores_i) > self.pos_nms_topk:inds = np.argsort(scores_i)[::-1]inds = inds[:self.pos_nms_topk]ret_i = ret_i[inds]rets.append(ret_i)return retsdef box_iou_xyxy(self, box1, box2):x1min, y1min, x1max, y1max = box1[0], box1[1], box1[2], box1[3]s1 = (y1max - y1min + 1.) * (x1max - x1min + 1.)x2min, y2min, x2max, y2max = box2[0], box2[1], box2[2], box2[3]s2 = (y2max - y2min + 1.) * (x2max - x2min + 1.)xmin = np.maximum(x1min, x2min)ymin = np.maximum(y1min, y2min)xmax = np.minimum(x1max, x2max)ymax = np.minimum(y1max, y2max)inter_h = np.maximum(ymax - ymin + 1., 0.)inter_w = np.maximum(xmax - xmin + 1., 0.)intersection = inter_h * inter_wunion = s1 + s2 - intersectioniou = intersection / unionreturn ioudef nms(self, bboxes, scores):# 对预测框得分进行排序inds = np.argsort(scores)inds = inds[::-1]keep_inds = []# 循环遍历预测框while(len(inds) > 0):cur_ind = inds[0]cur_score = scores[cur_ind]# 如果预测框得分低于阈值，则退出循环if cur_score < self.score_thresh:break# 计算当前预测框与保留列表中的预测框IOU，如果小于阈值则保留该预测框，否则丢弃该预测框keep = Truefor ind in keep_inds:current_box = bboxes[cur_ind]remain_box = bboxes[ind]# 计算当前预测框与保留列表中的预测框IOUiou = self.box_iou_xyxy(current_box, remain_box)if iou > self.nms_thresh:keep = Falsebreakif keep:keep_inds.append(cur_ind)inds = inds[1:]return np.array(keep_inds)

本文来自互联网用户投稿，文章观点仅代表作者本人，不代表本站立场，不承担相关法律责任。如若转载，请注明出处。 如若内容造成侵权/违法违规/事实不符，请点击【内容举报】进行投诉反馈！