Syn4D: A Multiview Synthetic 4D Dataset

Syn4D is a synthetic 4D dataset with multi-view RGB videos, depth, masks, tracking geometry, and supporting object mesh metadata.

Layout

data/
  syn4d_v1_stride_1/       # Syn4D V1 archives for stride-1 use
  syn4d_v1_stride_5/       # Syn4D V1 archives for stride-5 use
  syn4d_sim/               # simulation subset archives
  metadata/new_weight_bone/

code/visualizer/
  README.md
  requirements.txt
  run.py

code/bedlam2/
  README.md
  smplx_anim_to_alembic_batch.py
  smplx_anim_to_objs_batch.py

LICENSE
README.md
THIRD_PARTY_NOTICES.md
LICENSES/
resources/

Each Syn4D V1 split includes sequence_to_asset_mapping.csv, which maps dataset sequences to their underlying assets. The same mapping files are also referenced in THIRD_PARTY_NOTICES.md together with third-party asset notices and license references.

data/syn4d_v1_stride_1 and data/syn4d_v1_stride_5 contain Syn4D V1 packaged for different sampling strides. data/syn4d_sim contains the simulation subset. data/metadata contains shared object mesh and rig metadata used by the visualizer and dataset loaders.

The visualizer code is included under code/visualizer/.

Metadata note: We need the assets from BEDLAM2 to restore our tracking annotation. Processed BEDLAM2 data is not directly provided here due to BEDLAM2 license restrictions. Users can obtain BEDLAM2 from the official source and use the instructions under code/bedlam2/ to download the required assets and convert SMPL-X motion files into vertex-cache metadata. Please refer to code/bedlam2/README.md for further downloading instructions.

Performance note: The BEDLAM2 SMPL-X motion-to-vertex-cache conversion is CPU-heavy, so it is useful to run it on machines with fast CPUs.