def compute_metrics()

in threedod/benchmark_scripts/utils/eval_utils.py [0:0]

• 41 lines of code
• 16 McCabe index (conditional complexity)

    def compute_metrics(self):
        """Use accumulated predictions and groundtruths to compute Average Precision.
        # 1. calculate by-category precision, recall with multiprocessing mode
            each process for each category.

        Args:

        Returns:
            ret_dict: dict
                mAP: float (average AP over all categories)
                AR: float (average recall over all categories)
        """
        pred = {}
        gt = {}

        for img_id in pred_all.keys():
            for classname, bbox, score in pred_all[img_id]:
                if classname not in pred: pred[classname] = {}
                if img_id not in pred[classname]:
                    pred[classname][img_id] = []
                if classname not in gt: gt[classname] = {}
                if img_id not in gt[classname]:
                    gt[classname][img_id] = []
                pred[classname][img_id].append((bbox,score))
        
        for img_id in gt_all.keys():
            for classname, bbox in gt_all[img_id]:
                if classname not in gt: gt[classname] = {}
                if img_id not in gt[classname]:
                    gt[classname][img_id] = []
                gt[classname][img_id].append(bbox)
        
        rec = {}
        prec = {}
        ap = {}
        for classname in gt.keys():
            print('Computing AP for class: ', classname)
            rec[classname], prec[classname], ap[classname] = eval_det_cls(pred[classname], gt[classname], ovthresh, use_07_metric, get_iou_func)
            print(classname, ap[classname])

        ret_dict = {} 
        for key in sorted(ap.keys()):
            clsname = self.class2type_map[key] if self.class2type_map else str(key)
            ret_dict['%s Average Precision'%(clsname)] = ap[key]
        ret_dict['mAP'] = np.mean(list(ap.values()))
        rec_list = []
        for key in sorted(ap.keys()):
            clsname = self.class2type_map[key] if self.class2type_map else str(key)
            try:
                ret_dict['%s Recall'%(clsname)] = rec[key][-1]
                rec_list.append(rec[key][-1])
            except:
                ret_dict['%s Recall'%(clsname)] = 0
                rec_list.append(0)
        ret_dict['AR'] = np.mean(rec_list)
        return ret_dict