Created by: Edwardwaw

def __call__(self, sample, context=None):
    if 'cutmix' not in sample:
        return sample
    factor = np.random.beta(self.alpha, self.beta)
    factor = max(0.0, min(1.0, factor))
    if factor >= 1.0:
        sample.pop('cutmix')
        return sample
    if factor <= 0.0:
        return sample['cutmix']
    img1 = sample['image']
    img2 = sample['cutmix']['image']
    img = self._rand_bbox(img1, img2, factor)
    gt_bbox1 = sample['gt_bbox']
    gt_bbox2 = sample['cutmix']['gt_bbox']
    gt_bbox = np.concatenate((gt_bbox1, gt_bbox2), axis=0)
    gt_class1 = sample['gt_class']
    gt_class2 = sample['cutmix']['gt_class']
    gt_class = np.concatenate((gt_class1, gt_class2), axis=0)
    gt_score1 = sample['gt_score']
    gt_score2 = sample['cutmix']['gt_score']
    gt_score = np.concatenate(
        (gt_score1 * factor, gt_score2 * (1. - factor)), axis=0)
    sample['image'] = img
    sample['gt_bbox'] = gt_bbox
    sample['gt_score'] = gt_score
    sample['gt_class'] = gt_class
    sample['h'] = img.shape[0]
    sample['w'] = img.shape[1]
    sample.pop('cutmix')
    return sample

请问当前repo中这种gt box gt、label、 gt score的后处理方式在detection任务中是否是正确？是否要滤除掉因cut 而在最终增强的图像中未出现的gt box?