Skip to content

Problem in Open-loop Evaluation #60

@Alexanderofhust

Description

@Alexanderofhust

Great work!
However, I encountered environment issues when reproducing the Open-loop Evaluation:
My environment is configured with CUDA 12.2.Since PyTorch does not provide dedicated support for CUDA 12.2, the installation command I used is:
conda install pytorch==2.5.1 torchvision==0.20.1 torchaudio==2.5.1 pytorch-cuda=12.1 -c pytorch -c nvidia

This is consistent with the command provided in readme.md under the data_qa_generate/data_engine/datasets/nuscenes directory, and I followed all subsequent environment configuration steps strictly as outlined in the file.
But when running evaluation_nuscenes.py, I encountered the following error:
ModuleNotFoundError: No module named 'mmcv._ext'

After checking the MMCV official documentation, the root cause is that MMCV 1.7.2 installed via mim install does not support PyTorch > 2.0, while your provided code is built on MMCV < 2.0.
To resolve this, I reconfigured another environment with the following command to be compatible with MMCV 1.7.1:
conda install pytorch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1 cudatoolkit=11.7 -c pytorch -c conda-forge

However, running evaluation_nuscenes.py again resulted in this error:
ImportError: libcudart.so.12: cannot open shared object file: No such file or directory
This indicates that the module depends on CUDA > 12, but MMCV 1.x versions (per official docs) do not support CUDA > 12 (i.e., PyTorch > 2.0). It seems this environment conflict cannot be resolved.
Looking forward to your reply.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions