将COCO测试结果保存为HiEve格式

背景：

最近刚刚在HiEve数据集上做了实验，主要针对多人的姿态估计。因为dataloader一套主要参考COCO去做的，所以记录一下对测试结果格式的保存。将coco的测试结果转换为HiEve的格式，并保存成json。

---

修改方式

COCO结果保存的代码位于${project_root_dir}/lib/dataset/coco.py，主要代码是_coco_keypoint_results_one_category_kernel(self, data_pack)函数（大概位于359行左右）。将该函数进行了重写，如下所示：

def _coco_keypoint_results_one_category_kernel(self, data_pack):
        annolist = []

        keypoints = data_pack['keypoints']

        for img_kpts in keypoints:             
            frame_instance = {}
            frame_instance['ignore_regions'] = []

            annorect = []
            image = '%06d.jpg'%img_kpts[0]['image']
            frame_instance['image'] = [{'name': image}]
            for person in img_kpts:
                bbox_score = person['score']
                x1, y1, x2, y2 = person['bbox'][0], person['bbox'][1], person['bbox'][2], person['bbox'][3]
                annopoints = []
                point = []
                
                for i in range(person['keypoints'].shape[0]):
                    joint = {
                        'y': [person['keypoints'][i, 1]],
                        'score': [1],
                        'id': [i],
                        'x': [person['keypoints'][i, 0]]
                    }
                    point.append(joint)
                pose = {'point': point}
                annopoints.append(pose)

                instance = {
                    'score': [bbox_score],
                    'y1': [y1],
                    'y2': [y2],
                    'x2': [x2],
                    'x1': [x1],
                    'annopoints': annopoints,
                }
                annorect.append(instance)
            frame_instance['annorect'] = annorect
            annolist.append(frame_instance)
        hieve_res = {'annolist': annolist}      
        return hieve_res

---