search

metadriverse / trafficgen

[ICRA 2023] The official code for paper "TrafficGen: Learning to Generate Diverse and Realistic Traffic Scenarios"
Apache License 2.0
149 stars 24 forks source link

Some questions about data and training #25

Closed wjl2244 closed 1 year ago

wjl2244 commented 1 year ago

Hello, thank you very much for a meaningful job! I want to retrain this model, first I download all the 20s data from the https://console.cloud.google.com/storage/browser/waymo_open_dataset_motion_v_1_1_0/uncompressed/scenario/training_20s, it has a total of 50 compressed files, I process it through the sh utils/data_trans.sh PATH_A PATH_B script to produce a total of 3456 pkl files, However, you mentioned in the program comment that the number of files was 70542, may I ask if you only used waymo training_20s data?

Secondly, if I want to reproduce the results of the paper, do init and act models need to be trained separately? In the Cluster Training section of README.md, you only mentioned python train_act.py, do you need to run python train_init.py?

Looking forward to your reply and thanks again for your work!

Alan-LanFeng commented 1 year ago

Thanks for your question.

First: Yes, we only used 20s data. Second: You need to train two models separately.

LeonJoe13 commented 1 year ago

Hello, thank you very much for a meaningful job! I want to retrain this model, first I download all the 20s data from the https://console.cloud.google.com/storage/browser/waymo_open_dataset_motion_v_1_1_0/uncompressed/scenario/training_20s, it has a total of 50 compressed files, I process it through the sh utils/data_trans.sh PATH_A PATH_B script to produce a total of 3456 pkl files, However, you mentioned in the program comment that the number of files was 70542, may I ask if you only used waymo training_20s data?

Secondly, if I want to reproduce the results of the paper, do init and act models need to be trained separately? In the Cluster Training section of README.md, you only mentioned python train_act.py, do you need to run python train_init.py?

Looking forward to your reply and thanks again for your work!

It seems that you didn't download all training_20s data，maybe only 50