search

Henry1iu / TNT-Trajectory-Prediction

A Unofficial Pytorch Implementation of TNT: Target-driveN Trajectory Prediction
504 stars 95 forks source link

预训练模型性能问题 #31

Closed AIChuY closed 2 years ago

AIChuY commented 2 years ago

您好！我现在在试图用您提供的预训练权重重现模型，但是其表现似乎有些问题。在您提供的小型数据集的验证集上测试 python3 test_tnt.py -rm /media/wzy/78728a5f-27bc-43c6-ae04-0d7615fc9e86/wzy/TNT/TNT/best_TNT.pth -r ../interm_data_small -s val 得到的结果很差 The test result: {'minADE': 8.633853370351169, 'minFDE': 16.05028510884763, 'MR': 0.7764705882352941} 我自己作了可视化之后发现确实有一些很明显的错误 image 不知道您有没有头绪可能是什么造成这个结果？

AIChuY commented 2 years ago

似乎是包版本不一致导致的问题，在重新配置环境之后表现正常了 image

MozhganPourKeshavarz commented 1 year ago

Hi @AIChuY would you explain what was the problem? and have you tried an evaluation with the pre-trained weights on the actual test set?