smileyenot983/objaversexl_github6.5_pmap

NAMVIS training data, part 2

Multi-view renders used to train NAMVIS (NeurIPS 2026): GitHub objects of Objaverse-XL with an object-quality score above 6.5.

• 📄 Paper: arXiv:2610.04722
• 💻 Code: https://github.com/corl-team/namvis
• 🗂️ Both parts: Part 1 · Part 2

Format

290 WebDataset TAR shards (0.3 TB), named objaversexl_github-wNN-NNNNNN.tar; the first shard holds 170 objects. Each object (sample key = object ID) has 100 rendered views:

Member	Content
<key>.NNNN.webp	RGBA view, 256×256
<key>.NNNN.pointmap.exr	per-view point map
<key>.json	camera metadata: frames keyed by view, each with K (pixel intrinsics), camera_angle_x, depth_bounds, transform_matrix_original (Blender camera-to-world), and transform_matrix_normalized; plus the object's aabb

The training loader is infinity/dataset/webdataset_utils.py in the code repository. Download one shard to try it:

hf download smileyenot983/objaversexl_github6.5_pmap objaversexl_github-w00-000000.tar --repo-type dataset --local-dir data

License

The renders are derived from Objaverse-XL objects, each of which keeps its own license; see Objaverse-XL.

Citation

@inproceedings{khafizov2026namvis,
  title         = {{NAMVIS}: Next-Scale Autoregressive Multi-View Image Synthesis},
  author        = {Khafizov, Ramil and Statsenko, Ilya and Rakhimov, Ruslan and Komarichev, Artem and Wonka, Peter and Burnaev, Evgeny},
  booktitle     = {Advances in Neural Information Processing Systems},
  year          = {2026},
  eprint        = {2610.04722},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CV},
  url           = {https://arxiv.org/abs/2610.04722}
}