OpenDriveLab / ViDAR

[CVPR 2024 Highlight] Visual Point Cloud Forecasting
https://arxiv.org/abs/2312.17655
Apache License 2.0
266 stars 17 forks source link

How to use all the data to fine-tune? #14

Closed LuPaoPao closed 6 months ago

LuPaoPao commented 6 months ago

I looked at the configuration parameters and there didn't seem to be any parameters to control the amount of data. How do I fine-tune with all the data sets?

tomztyang commented 6 months ago

Here. Change it to 1 for full set pre-training.

For fine-tuning, cast the parameter of use_pretrain_data to False.

Thanks, Zetong

LuPaoPao commented 6 months ago

Here. Change it to 1 for full set pre-training.

For fine-tuning, cast the parameter of use_pretrain_data to False.

Thanks, Zetong

Thanks for reply. In "ViDAR-RN101-nus-1-8-1future", "1-8" means use 12.5% dataset? In your fine-tuning config "projects/configs/vidar_finetune/nusc_1_4_subset/vidar_1_8_nusc_1future.py", why use_pretrain_data=True?

Thank you again for your excellent work.

LuPaoPao commented 6 months ago

Here. Change it to 1 for full set pre-training. For fine-tuning, cast the parameter of use_pretrain_data to False. Thanks, Zetong

Thanks for reply. In "ViDAR-RN101-nus-1-8-1future", "1-8" means use 12.5% dataset? In your fine-tuning config "projects/configs/vidar_finetune/nusc_1_4_subset/vidar_1_8_nusc_1future.py", why use_pretrain_data=True?

Thank you again for your excellent work.

I have found the answer myself, and thank you very much for your excellent contribution. Thanks again.