search

Thinklab-SJTU / Bench2Drive

[NeurIPS 2024 Datasets and Benchmarks Track] Closed-Loop E2E-AD Benchmark Enhanced by World Model RL Expert
Apache License 2.0
1.24k stars 79 forks source link

mmcv version issue for closed-loop evaluation #7

Closed Fengtao22 closed 3 months ago

Fengtao22 commented 3 months ago

Hi, thanks for providing the closed-loop evaluation platform. One question with respect to the mmcv package: I tried to directly test VAD repo using the Bench2Drive platform, however, the VAD requries mmdet (2.14.0) which needs a different version of mmcv. Hence, there are some incompatibility issues.

For example, the mmdet checks mmcv version in its init.py will give "AssertionError: MMCV==0.0.1 is used but incompatible. Please install mmcv>=1.3.8, <=1.4.0." Even when I circulate this assert, I had another coming issue: AttributeError: module 'mmcv' has no attribute 'jit'.

Can you kindly provide some suggestions? Thanks!

Fengtao22 commented 3 months ago

I just realized that the mmdet, mmcv and etc are merged into the mmcv folder and I should use the new config instead of the original VAD config. However, I still have issues in running the leaderboard. The current problem is:

in _lazy_init torch._C._cuda_init() RuntimeError: No CUDA GPUs are available

jayyoung0802 commented 3 months ago

Hi, thanks for your attention. In order to remove the restrictions on torch and cuda versions in the original uniad/vad repository, we have refactored the uniad/vad code. The repo is Bench2DriveZoo. Please follow this link to install and evaluate.

Fengtao22 commented 3 months ago

Hi, thanks for your attention. In order to remove the restrictions on torch and cuda versions in the original uniad/vad repository, we have refactored the uniad/vad code. The repo is Bench2DriveZoo. Please follow this link to install and evaluate.

Thanks for your quick reply! I actually did follow the Bench2DriveZoo to set up my environment. However, I still have the issue with GPUs. Any clue?

Fengtao22 commented 3 months ago

Hi, thanks for your attention. In order to remove the restrictions on torch and cuda versions in the original uniad/vad repository, we have refactored the uniad/vad code. The repo is Bench2DriveZoo. Please follow this link to install and evaluate.

Thanks for your quick reply! I actually did follow the Bench2DriveZoo to set up my environment. However, I still have the issue with GPUs. Any clue?

It turns out that I was using a wrong GPU_RANK for the bash file and after I changed it to 0 (for single GPU setting) it finally works!!!