autonomousvision / transfuser

[PAMI'23] TransFuser: Imitation with Transformer-Based Sensor Fusion for Autonomous Driving; [CVPR'21] Multi-Modal Fusion Transformer for End-to-End Autonomous Driving
MIT License
1.12k stars 186 forks source link

CARLA setup error and Running CARLA server without display for evaluation #212

Closed SRajasekar333 closed 4 months ago

SRajasekar333 commented 4 months ago

Hi, I am facing some doubts while trying to evaluate the model, so please guide. I trained the transfuser model by running CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6 OMP_NUM_THREADS=48 OPENBLAS_NUM_THREADS=1 torchrun --nnodes=1 --nproc_per_node=7 --max_restarts=0 --rdzv_id=1234576890 --rdzv_backend=c10d train.py --logdir log --root_dir data --parallel_training 1 --epochs 32 --batch_size 48 --id transfuser and saved the 32 model_n.pth and optimizer_n.pth

So now I'm trying to evaluate the model as in README.md,

  1. When I setup and install the CARLA 0.9.10.1, I face the following error ERROR 404: Not Found

image

  1. I'm trying to evaluate the model using the HPC remote server, so in this case how to run the CARLA server without display?

    Can I run the server for Without Display - Without Docker condition as, SDL_VIDEODRIVER=offscreen SDL_HINT_CUDA_DEVICE=0 ./CarlaUE4.sh --world-port=2000 -opengl and then evaluate the agent using, ./leaderboard/scripts/local_evaluation.sh <carla root> <working directory of this repo (*/transfuser/)>

    Please guide on how to evaluate the model and save the output files instead of stream display.

Thanks in advance.

Regards,

Kait0 commented 4 months ago
  1. The download link for carla changed. You can use this script to download it (will update later).

  2. If the carla server and the evaluation script are on the same machine then yes, otherwise you need to set the ip and port etc.

Kait0 commented 4 months ago

I updated the links in the download scripts.