How to Calcualte PredCLS and SGCls ? about reltr HOT 4 CLOSED

def evaluate_batch_predcls(outputs, targets, evaluator, matching_indices, evaluator_list):

#TODO
for batch, target in enumerate(targets):
    target_bboxes_scaled = rescale_bboxes(target['boxes'].cpu(), torch.flip(target['orig_size'],dims=[0]).cpu()).clone().numpy() # recovered boxes with original size

    gt_entry = {'gt_classes': target['labels'].cpu().clone().numpy(),
                'gt_relations': target['rel_annotations'].cpu().clone().numpy(),
                'gt_boxes': target_bboxes_scaled}

    index1, index2 = matching_indices[1][batch]

    sub_bboxes_scaled = rescale_bboxes(outputs['sub_boxes'][batch].cpu(), torch.flip(target['orig_size'],dims=[0]).cpu()).clone().numpy()
    obj_bboxes_scaled = rescale_bboxes(outputs['obj_boxes'][batch].cpu(), torch.flip(target['orig_size'],dims=[0]).cpu()).clone().numpy()

    pred_sub_scores, pred_sub_classes = torch.max(outputs['sub_logits'][batch].softmax(-1)[:, :-1], dim=1)
    pred_obj_scores, pred_obj_classes = torch.max(outputs['obj_logits'][batch].softmax(-1)[:, :-1], dim=1)
    rel_scores = outputs['rel_logits'][batch][:,1:-1].softmax(-1)

    pred_sub_classes = pred_sub_classes.cpu().clone().numpy()
    pred_sub_scores = pred_sub_scores.cpu().clone().numpy()
    pred_obj_classes = pred_obj_classes.cpu().clone().numpy()
    pred_obj_scores = pred_obj_scores.cpu().clone().numpy()

    if len(gt_entry['gt_relations']) > 1:
        sub_bboxes_scaled[index1] = gt_entry['gt_boxes'][gt_entry['gt_relations'][index2][:, 0]]
        obj_bboxes_scaled[index1] = gt_entry['gt_boxes'][gt_entry['gt_relations'][index2][:, 1]]
        pred_sub_classes[index1] = gt_entry['gt_classes'][gt_entry['gt_relations'][index2][:, 0]]
        pred_obj_classes[index1] = gt_entry['gt_classes'][gt_entry['gt_relations'][index2][:, 1]]
        pred_sub_scores[index1] = 1
        pred_obj_scores[index1] = 1
    else:
        sub_bboxes_scaled[index1] = gt_entry['gt_boxes'][gt_entry['gt_relations'][index2][0]]
        obj_bboxes_scaled[index1] = gt_entry['gt_boxes'][gt_entry['gt_relations'][index2][1]]
        pred_sub_classes[index1] = gt_entry['gt_classes'][gt_entry['gt_relations'][index2][0]]
        pred_obj_classes[index1] = gt_entry['gt_classes'][gt_entry['gt_relations'][index2][1]]
        pred_sub_scores[index1] = 1
        pred_obj_scores[index1] = 1

    mask = (pred_sub_classes - pred_obj_classes != 0)
    if mask.sum() <= 198:
        sub_bboxes_scaled = sub_bboxes_scaled[mask]
        pred_sub_classes = pred_sub_classes[mask]
        pred_sub_scores = pred_sub_scores[mask]
        obj_bboxes_scaled = obj_bboxes_scaled[mask]
        pred_obj_classes = pred_obj_classes[mask]
        pred_obj_scores = pred_obj_scores[mask]
        rel_scores = rel_scores[mask]

    pred_entry = {'sub_boxes': sub_bboxes_scaled,
                  'sub_classes': pred_sub_classes,
                  'sub_scores': pred_sub_scores,
                  'obj_boxes': obj_bboxes_scaled,
                  'obj_classes': pred_obj_classes,
                  'obj_scores': pred_obj_scores,
                  'rel_scores': rel_scores.cpu().clone().numpy()}

    evaluator['predcls'].evaluate_scene_graph_entry(gt_entry, pred_entry)

    if evaluator_list is not None:
        for pred_id, _, evaluator_rel in evaluator_list:
            gt_entry_rel = gt_entry.copy()
            mask = np.in1d(gt_entry_rel['gt_relations'][:, -1], pred_id)
            gt_entry_rel['gt_relations'] = gt_entry_rel['gt_relations'][mask, :]
            if gt_entry_rel['gt_relations'].shape[0] == 0:
                continue
            evaluator_rel['predcls'].evaluate_scene_graph_entry(gt_entry_rel, pred_entry)