Open PuJanhan opened 3 years ago
Hi @PuJanhan, Thanks for the interest.
The printout from the scripts are at the point level. E.g. the recall of car points gets detected (thres=0.5). To get the object level mAP, use submodule kitti_native_evaluation:
Install kitti_native_evaluation offline evaluation:
cd kitti_native_evaluation
cmake ./
make
Evaluate output results on the validation split:
evaluate_object_offline DATASET_ROOT_DIR/labels/training/label_2/ DIR_TO_SAVE_RESULTS
It will give the easy moderate hard category as well.
I use python3 train.py configs/car_auto_T3_train_train_config configs/car_auto_T3_train_config and python3 eval.py configs/car_auto_T3_train_eval_config, they get the results like this : Class_0: recall=0.992798, prec=0.992376, mAP=0.999531, loc=0.000000 x=0.0000 y=0.0000 z=0.0000 l=0.0000 h=0.0000 w=0.0000 y=0.0000 Class_1: recall=0.807884, prec=0.801547, mAP=0.872123, loc=7.071270 x=0.7826 y=0.5973 z=2.0156 l=0.9511 h=0.5748 w=0.6402 y=1.5098 Class_2: recall=0.838907, prec=0.830165, mAP=0.911929, loc=19.158129 x=0.9189 y=1.2436 z=8.2396 l=2.5387 h=1.5434 w=1.7170 y=2.9570 Class_3: recall=0.783943, prec=0.834315, mAP=0.873242, loc=0.000000 x=0.0000 y=0.0000 z=0.0000 l=0.0000 h=0.0000 w=0.0000 y=0.0000 but in your paper ,three types of car such as easy moderate hard , they has their own map, i dont konw where get the results of paper in your code and the dataset dont have the three types . In addition,what the correspondence between the output of train and the results in your paper. which result in the code represent the result in your paper
Hi,Did you verify the model successfully?
Sorry,I don't have time and equipment to do it recently. I will try it later. thanks for your reply------------------ 原始邮件 ------------------ 发件人: "typhoonlee"<notifications@github.com> 发送时间: 2021年1月18日(星期一) 上午10:10 收件人: "WeijingShi/Point-GNN"<Point-GNN@noreply.github.com>; 抄送: "PuJH"<805280381@qq.com>;"Mention"<mention@noreply.github.com>; 主题: Re: [WeijingShi/Point-GNN] The output of the train and eval (#65)
Yes ,I follow the advice of https://github.com/WeijingShi/Point-GNN/issues/69and success
I use python3 train.py configs/car_auto_T3_train_train_config configs/car_auto_T3_train_config and python3 eval.py configs/car_auto_T3_train_eval_config, they get the results like this : Class_0: recall=0.992798, prec=0.992376, mAP=0.999531, loc=0.000000 x=0.0000 y=0.0000 z=0.0000 l=0.0000 h=0.0000 w=0.0000 y=0.0000 Class_1: recall=0.807884, prec=0.801547, mAP=0.872123, loc=7.071270 x=0.7826 y=0.5973 z=2.0156 l=0.9511 h=0.5748 w=0.6402 y=1.5098 Class_2: recall=0.838907, prec=0.830165, mAP=0.911929, loc=19.158129 x=0.9189 y=1.2436 z=8.2396 l=2.5387 h=1.5434 w=1.7170 y=2.9570 Class_3: recall=0.783943, prec=0.834315, mAP=0.873242, loc=0.000000 x=0.0000 y=0.0000 z=0.0000 l=0.0000 h=0.0000 w=0.0000 y=0.0000 but in your paper ,three types of car such as easy moderate hard , they has their own map, i dont konw where get the results of paper in your code and the dataset dont have the three types . In addition,what the correspondence between the output of train and the results in your paper. which result in the code represent the result in your paper