Preprocessed kitti for DVGT training (MoGe-v2 annotations), mirroring yiren-lu/DVGT-navsim. Splits: val train.
Contents: image / proj_depth / align_depth under data_annotation//moge_v2_large_correct_focal/kitti/, plus meta parquets.
cat data_annotation_kitti.tar.part-* | tar -xf -
