search

bradyz / cross_view_transformers

Cross-view Transformers for real-time Map-view Semantic Segmentation (CVPR 2022 Oral)
MIT License
531 stars 81 forks source link

question about eval size and preprocess #11

Open JumpXing opened 2 years ago

JumpXing commented 2 years ago

Hi，What size images did you use when evaluating, still 480x224 or origin size of 1600x900

JumpXing commented 2 years ago

And you rotated and translate the camera image before training, but you only adjust the camera intrinsics and didn't adjust the camera extrinsics. I don't know if this is right.

bradyz commented 2 years ago

the camera image is only resized and cropped during training which only affects the camera intrinsics for all stages, both train and eval, we use 480x224