score = list()
for i in range(1, 1 + self.n_class):
mask = raw_score[:, i] >= self.score_threshold
bbox_label, score_label = raw_bbox[mask], raw_score[mask][:, i]
if self.nms_threshold is not None:
order = xp.array(chainer.cuda.to_cpu(
score_label).argsort()[::-1])
After Change
xp = self.xp
raw_bbox = chainer.cuda.to_cpu(raw_bbox)
raw_score = chainer.cuda.to_cpu(raw_score)
bbox = list()
label = list()
score = list()
for i in range(1, 1 + self.n_class):
mask = raw_score[:, i] >= self.score_threshold
bbox_label = raw_bbox[mask]
score_label = raw_score[mask, i]
if self.nms_threshold is not None:
order = score_label.argsort()[::-1]
bbox_label, score_label = bbox_label[order], score_label[order]
bbox_label, param = transforms.non_maximum_suppression(
bbox_label, self.nms_threshold, return_param=True)
score_label = score_label[param["selection"]]
