some questions on testing opendrivevla model on other vqa benchmarks

[krishnakanthnakka]

Hi Team,

Thank you for opensourcing the codebase. I wanted to ask a question if I could use the released checkpoint on test on vqa benchmarks which contains image prompt and text prompt (containing question + choices) without any additional metadata information like steering, historical trajectory, speed, acceleration, etc.

If the input information is just simple vision image input and text prompt, can i adapt your inference pipeline code with the released checkpoint.

Thank you for answering.

[TheNeeloy]

Hey! I just realized after posting my issue that it's related to yours: #31
It seems like the inference can work on other datasets and simulators once we remove ground truth data, albeit with some changes to the predicted trajectory.
Let me know if you make any progress in this direction as well; thanks!