Preprocessed nuScenes 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/nuscene/, plus meta parquets (raw + debug1 filtered).
cat data_annotation_nuscene.tar.part-* | tar -xf -
