因为在深度学习中的目标检测中会检测出多个目标框,后期需要通过非极大值抑制去除得分低并且iou大于阈值的目标框。
因此,在此我们实现了一个简单的nms的python程序,以此作为记录。
nms代码:
# --*-- coding:utf8 --*-- import operator import numpy as np def iou(box1, box2): x1, y1, w1, h1, s1 = box1 x2, y2, w2, h2, s2 = box2 xl = max(x1, x2) yl = max(y1, y2) xr = min(x1+w1, x2+w2) yr = min(y1+h1, y2+h2) area = (xr-xl)*(yr-yl) return float(area)/float(w1*h1+w2*h2-area) def nms(boxes, thresh): boxes.sort(key=operator.itemgetter(4), reverse=True) c = 1 while c<len(boxes): box = boxes[c-1] i = c while i<len(boxes): if iou(box, boxes[i]) >= thresh: del boxes[i] i -= 1 i += 1 c += 1 return boxes loc = np.random.randint(1, 50, [50, 2]).tolist() size = np.random.randint(50, 100, [50, 2]).tolist() score = [(x+0.4)/1.4 for x in np.random.rand(50)] record = [] for i in range(50): record.append(loc[i]+size[i]+[score[i]]) boxes = nms(record, 0.3) print(boxes)