Generation and training on Carla Dataset

[ZHDphilip]

Hi, thanks for this great piece of work.
Currently, I am working on training the network using the carla dataset, and when trying to generate videos using generate.py, I found that the video is not a 360-degree view around the car, instead the video focuses on a small range of views, no matter how I try to modify the camera trajectory. When modifying camera trajectory, I changed the relative_range_u arg, I want to make sure if I am modifying the correct argument. I have been suspecting that I need a new set of camera kwargs for Carla, so I want to first assure that the error is not arisen from a wrong rendering trajectory. Thanks

The training command I used:
python run_train.py outdir={path} data={path} spec=paper256 model=stylenerf_cars