Datasets:

toyxyz
/

booru3dpose_10k

Exception:    SplitsNotFoundError
Message:      The split names could not be parsed from the dataset config.
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
                  for split_generator in builder._split_generators(
                                         ~~~~~~~~~~~~~~~~~~~~~~~~~^
                      StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 80, in _split_generators
                  raise ValueError(
                  ...<2 lines>...
                  )
              ValueError: The TAR archives of the dataset should be in WebDataset format, but the files in the archive don't share the same prefix or the same types.
              
              The above exception was the direct cause of the following exception:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 68, in compute_split_names_from_streaming_response
                  for split in get_dataset_split_names(
                               ~~~~~~~~~~~~~~~~~~~~~~~^
                      path=dataset,
                      ^^^^^^^^^^^^^
                      config_name=config,
                      ^^^^^^^^^^^^^^^^^^^
                      token=hf_token,
                      ^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
                  info = get_dataset_config_info(
                      path,
                  ...<6 lines>...
                      **config_kwargs,
                  )
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
                  raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
              datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

Booru3DPose 10K

Paired anime images, shaded 3D body-mesh conditioning images with fingers, and captions for training an Anima ControlNet-LLLite pose adapter. Each condition matches its target's dimensions and framing. Meshes have a black background, a different randomly varied solid material color per person, and camera-relative lighting that makes body shape and overlapping parts visible. Expressions are neutralized.

Split	Pairs
Train	10,080
Validation	10
Test	10
Total	10,100

The training split contains 6,220 Anima-generated images and 3,860 original images from Curated Danbooru 2026. Existing generated-image captions are retained; original-image captions use the source dataset's stored tags without added prose. Both holdout splits contain generated images and remain separate from training.

Training-sized revision

Targets and mesh conditions use at most 1,048,576 pixels, a 1536px longest edge, and 16-pixel-aligned canvases. Aspect ratio is retained, with no upscaling or cropping and identical minimal centered black padding for both images. 1,903 training pairs were resized; all 10,100 pairs and both holdout splits are retained.

The original-resolution release remains accessible at revision c75214bb7370c7da189fc30a5adef60ec444f95e. See resizing and step accounting.

Download and prepare

Python 3.11+ and the Hugging Face CLI are required. This prepares data only:

hf download toyxyz/booru3dpose_10k --repo-type dataset --local-dir /workspace/booru3dpose_release
python /workspace/booru3dpose_release/prepare_training.py \
  --output /workspace/booru3dpose_10k --batch-size 32 --caption-dropout 0.15

The script verifies archives and all extracted files against SHA256, then generates native sd-scripts dataset TOMLs. Use dataset_train.toml for training; validation and test configurations are separate. dataset.toml is a train-only alias. The command selects batch 32 and train caption dropout 0.15 for the planned 96GB Anima setup. Use a smaller batch on GPUs with less memory; validation and test configs use batch 1 and zero dropout. The script never starts training.

The release uses 9 uncompressed TAR shards, approximately 13.60 GB / 12.67 GiB, plus manifests and documentation. Allow at least 35 GiB free space for download and extraction. Files are grouped by their actual image dimensions. This is the same portable image/TXT training-package format as BooruPose 10K.

See training and layout, sources and mesh preparation, and release metadata.

Review

All retained pairs passed completed manual pose review; flagged pairs were removed. PNG text/EXIF metadata is removed. Exact caption bytes are preserved. This training-sized revision downsamples 1,903 pairs; unchanged pairs retain their previously published pixel data. Pose estimation can still fail on occlusion, cropped hands, unusual anatomy or multiple people. These are estimated illustration poses, not motion-capture ground truth. Content rating was not filtered in the original-image additions; the source collection can contain mature or explicit artwork. See source notes for rights and model terms.

Downloads last month: 166

Total file size:

13.6 GB

Models trained or fine-tuned on toyxyz/booru3dpose_10k

toyxyz/Anima_ControlNet-LLLite

Image-to-Image • Updated 14 days ago • 29