About some problems that arise during training

[water221]

My system is Ubantu 20.04.1, and the training environment is GeForce RTX3090, python 3.10, CUDA=11.7. I used the command python train.py--config config/train_marigold.yaml --no_wandbto start the project and it has been stuck here without any follow-up actions and no log output. Does anyone know where the problem is? Can you help solve it? In addition, I use Hypersim and vkitti datasets

[water221]

Supplement:Only the above log output, and then there is no other output or action

[YacineDeghaies]

Did you use Conda for the dependencies or pip ?

[water221]

Did you use Conda for the dependencies or pip ?

Yes, I am using the dependency environment for conda installation. After investigation, I now feel that the dataset has not been downloaded completely. I wonder if anyone can answer whether the evaluation dataset also needs to be downloaded completely?

[YacineDeghaies]

Can you please share your dependencies for training ? I'm also doing the same trying to train it.

[water221]

Can you please share your dependencies for training ? I'm also doing the same trying to train it.

You can find the dependencies you want in the three files: requirements.txt, requirements+.txt and requirements++.txt. I installed them using the pip command

[YacineDeghaies]

Where you able to train the model ?

[water221]

Where you able to train the model ?

On my server

[YacineDeghaies]

培训对你有用吗？

[markkua]

I have never encountered this. Some people found the uncompressed tar file to be the obstacle https://github.com/prs-eth/Marigold/issues/94.