how to evaluate the map using coco metrix in custom dataset?

[luoshiyong]

``i had implement a custom dataset in instance segmentation ,which labels is labelme json, but in mmdet ,there is no evaluation metric for labelme json, so i modified the coco metrix as follows: val_evaluator = dict( type='CisdiMetric', ann_file=None, metric=['bbox', 'segm'], format_only=False, backend_args=backend_args) test_evaluator = val_evaluator , and in metrics, i use the following code to get the `gt,if self._coco_api is None:
# TODO: Need to refactor to support LoadAnnotations
assert 'gt_instances' in data_sample,
'ground truth is required for evaluation when '
'`ann_file` is not provided'
gt['anns'] = []
# print(data_sample)
boxes = data_sample['gt_instances']['bboxes'].detach().cpu().numpy()
labels = data_sample['gt_instances']['labels'].detach().cpu().numpy()
masks = [] if data_sample['gt_instances'].get('masks',None) is None else data_sample['gt_instances']['masks'].raw_masks
# print(data_sample['gt_instances']['masks'])
for bbox, label, mask in zip(boxes, labels, masks):
gt['anns'].append({'bbox': bbox, 'bbox_label': label,"mask":mask})
# gt['anns'] = data_sample['gt_instances']
# add converted result to the results list
self.results.append((gt, result))`, but the result is all map is 0,

[luoshiyong]