python-cpu-nms实现和python-cpu-merge-nms实现
python-cpu-nms
nms非极大值抑制处理
def py_cpu_nms(dets, nms_thres, nms_top_k=300, score_threshold=0.1):
"""
Pure Python NMS baseline.
nms_top_k (int): Maximum number of detections to be kept according to
the confidences after the filtering detections based
on score_threshold.
nms_threshold (float): The threshold to be used in NMS. Default: 0.3
"""
# 筛除得分较低的检测框
dets = dets[dets[:, 4] > score_threshold]
x1 = dets[:, 0]
y1 = dets[:, 1]
x2 = dets[:, 2]
y2 = dets[:, 3]
scores = dets[:, 4]
areas = (x2 - x1 + 1) * (y2 - y1 + 1)
order = scores.argsort()[::-1][:nms_top_k]
keep = []
while order.size > 0:
i = order[0]
keep.append(i)
xx1 = np.maximum(x1[i], x1[order[1:]])
yy1 = np.maximum(y1[i], y1[order[1:]])
xx2 = np.minimum(x2[i], x2[order[1:]])
yy2 = np.minimum(y2[i], y2[order[1:]])
w = np.maximum(0.0, xx2 - xx1 + 1)
h = np.maximum(0.0, yy2 - yy1 + 1)
inter = w * h
ovr = inter / (areas[i] + areas[order[1:]] - inter)
inds = np.where(ovr <= nms_thres)[0]
order = order[inds + 1]
# print(dets[keep])
return dets[keep]
python-cpu-merge-nms
对于一批重复率较高的框不是简单的去置信度最高的预测框,而且根据置信度赋予每个预测框一个权重值,置信度较高权重也较高,因为置信度高有理由更加的看重这个预测框。所以,对于所以的预测框乘上一个置信度权重,简单来说就是对预测框信息做一个权重和取平均,思想上是可以重合每个边界框的信息。主要作用是优化定位。
def py_cpu_merge_nms(dets, nms_thres, nms_top_k=300, score_threshold=0.1, merge_threshold=0.9):
# merge_threshold参数:大于这个iou的框合并在一起
# 筛除得分较低的检测框
dets = dets[dets[:, 4] > score_threshold]
x1 = dets[:, 0]
y1 = dets[:, 1]
x2 = dets[:, 2]
y2 = dets[:, 3]
scores = dets[:, 4]
areas = (x2 - x1 + 1) * (y2 - y1 + 1)
order = scores.argsort()[::-1][:nms_top_k]
keep = []
while order.size > 0:
i = order[0]
xx1 = np.maximum(x1[i], x1[order[1:]])
yy1 = np.maximum(y1[i], y1[order[1:]])
xx2 = np.minimum(x2[i], x2[order[1:]])
yy2 = np.minimum(y2[i], y2[order[1:]])
w = np.maximum(0.0, xx2 - xx1 + 1)
h = np.maximum(0.0, yy2 - yy1 + 1)
inter = w * h
ovr = inter / (areas[i] + areas[order[1:]] - inter)
inds = np.where(ovr <= nms_thres)[0]
merge_ind = np.where(ovr > merge_threshold)[0]
# 乘以权重(得分也取了权重 可能受得分低的框 整体低一些)
weights = dets[order][merge_ind+1, 4:5]
keep.append((dets[order][merge_ind+1] * weights).sum(0) / weights.sum())
# 不乘以权重直接取平均 (得分我取的最高)
# if len(merge_ind):
# keep.append(np.insert(dets[order][merge_ind+1][:,:4].mean(axis=0),4,dets[order][merge_ind[0]+1][4]))
order = order[inds + 1]
keep = np.array(keep)
# print(keep)
return keep