The Err is nan

[raozhongyu]

Hi, thanks for your great work. When I use my own dataset which collected with the carla simulation and the baseline is 50 CM. I meet the err is nan, Can you give me some suggestion. I use the deafult setting. Thanks a lot.

[fabiotosi92]

Hi! Is it a problem that arises at the beginning of the training?
Let's try to change the eps to a smaller value (e.g. 1e^-3) value in "lib/model/SMDHead.py".

[raozhongyu]

Yes， thanks a lot. I also want to konw if this code can be trained on multi-gpu.

[fabiotosi92]

We trained our models using a single GPU but the code can be adapted to use multiple GPUs.

[raozhongyu]

Thanks a lot