What about the GPU usage when finetuning?

[dingli-dean]

First of all, thx for releasing this excellent work. In general, the usage of GPU in finetuning stage is obviously less than that in pre-train stage. However, I try to excute finetuning based on the pretrained model (trained with mem_efficient_vidar_1_8_nusc_3future_r50.py), the GPU usage reaches more than 40G. Thus, can you inform the GPU usage when finetuning?

Thx.

[tomztyang]

Hi,

Thanks for your interest.

During fine-tuning, we include the pre-trained latent rendering module, which results in the computational cost, as shown in here.

You can also refer to our fine-tuning logs at here. It costs around 40G for fine-tuning.

Thanks, Zetong

[tomztyang]

I will close this issue for now, feel free to re-open it for further questions.

Thanks, Zetong