Inference on Live Frames from CARLA

[Ali-Majlesi]

Hi I am trying to connect BEVFormer to the CARLA Simulator. My goal is to capture images from 6 cameras in CARLA and pass them directly to BEVFormer for inference. I found how to perform inference on a dataset stored in a folder, but I couldn’t find a way to perform inference on 6 images in the form of TensorFlow tensors (instead of files in a directory). Essentially, I want to perform inference on single frames consecutively from a driving video in CARLA.

[Rushikesh22750]

Hi Ali,

Did you found any way out ? Please help me also.

Thankyou, Rushikesh

[Ali-Majlesi]

Hi Ali,

Did you found any way out ? Please help me also.

Thankyou, Rushikesh

@Rushikesh22750 Hi, I would like to discuss this in more detail. Can you share an email address where I can reach you? Or you can email me at [email protected]

[Rushikesh22750]

Hi Ali, You can contact me at [email protected]

[medem23]

I'm also quite struggling understanding a way to do this, did you figure this out ?

[artbelyaev0]

@Ali-Majlesi Hi! Can you show me please a minimal example for running BEVFormer with real images ?

[yaobaishen]

@Ali-Majlesi Hi! Can you show me please a minimal example for running BEVFormer with real images ?

@artbelyaev0 hihi, I am looking for the solution to run on real images too (even poor performance maybe expected), do you have any findings? thanks a lot.

[QZJGeorge]

Hello, we are the authors of nuCarla. We have successfully adapted BEVFormer in the CARLA simulator and provided several pretrained backbones. We welcome you to take a look!