The eval code for the calculation of ap is wrong #10
Description
In the eval.py file, pred_boxes,pred_classes,pred_scores are sorted before calculating ap. In eval_ap_2d function, zip(gt_single_cls, bbox_single_cls, scores_single_cls) was used. but the sizes of gt_single_cls, bbox_single_cls, scores_single_cls are different. This will make low-score results ignored. is it reasonable？ eg: gt_single_cls=[[],[3],[],[3]], bbox_single_cls=[[],[3],[3],[3],[3]], scores_single_cls=[[],[0.9],[0.8],[0.7],[0.6]], the 0.6 box will be ignored. but maybe 0.9,0.8,0.7 has the same gt. Consequently， it will cause lower ap.