DerrickXuNu/OpenCOOD — reverse-engineered prompt

I want this repo turned into a working, easy to run cooperative driving perception project on my machine. Please get the environment set up, make sure the data flow works for OPV2V and V2XSet if supported, and keep the main focus on LiDAR since that looks like the most complete path here.

I should be able to download or point to the dataset, visualize a data sequence, train one of the included detection models from a config file, and run inference on a saved checkpoint with the different fusion options. If possible, also make the log replay tool usable for OPV2V. Multi GPU training support would be great if the project already has it.

Please wire up anything missing, fix broken paths or config issues, and leave me with a simple set of commands that actually run end to end. A short getting started guide in the repo would help too, including where to put the data and how to test that everything is working. You can check current docs online if needed.