Thanks for your great job! That was amazing to see the e2e planning works successfully in multi-view input scenarios.
About your CARLA experiment, I wonder if you conduct any resizing before cropping. In stp3/datas
/CarlaData.py, the original image size was set to (300, 400) and directly cropped into (256, 256) without any compression. Limited FOV and direct cropping mean some parts of views would be uncovered in the image inputs. Do you have any scheme to process this issue, or have you tried to resize the images for CARLA experiments?
Thanks for your great job! That was amazing to see the e2e planning works successfully in multi-view input scenarios.
About your CARLA experiment, I wonder if you conduct any resizing before cropping. In stp3/datas /CarlaData.py, the original image size was set to (300, 400) and directly cropped into (256, 256) without any compression. Limited FOV and direct cropping mean some parts of views would be uncovered in the image inputs. Do you have any scheme to process this issue, or have you tried to resize the images for CARLA experiments?
谢谢~ Best regards, Chi-Chang Lee