question about training details

[booker-max]

1. The implementation details are shown in the figure, but I am a bit confused about the crop_size setting of
   [450, 600] during the training of the outdoor model.

2. Additionally, I would like to ask about the specific meaning of ["model"]["pixel_encoder"]["img_size"] in the config file. During training, this variable is set to [500, 700]，Does this mean it remains [500, 700] throughout the entire training process?

Furthermore, how should this variable be set during testing? I plan to test on my own dataset, where the img_size settings for different datasets are shown in the figure below.

3. Looking forward to the release of a foundation-level model trained on a large-scale, diverse dataset mixture, covering perspective, fisheye, and 360-degree camera data.