Datasets:

initialneil
/

TEDUB10k-preview

license: cc-by-nc-nd-4.0
task_categories:
  - image-to-3d
tags:
  - human
  - avatar
  - SMPL-X
  - FLAME
  - MANO
  - TED
  - upper-body
size_categories:
  - 1K<n<10K
configs:
  - config_name: default
    data_files:
      - split: train
        path: metadata/subjects_train.parquet
      - split: val
        path: metadata/subjects_val.parquet
      - split: train_subset_x1
        path: metadata/subjects_train_subset_x1.parquet
      - split: train_subset_x12
        path: metadata/subjects_train_subset_x12.parquet
      - split: train_val
        path: metadata/subjects_train_val.parquet

TEDUB10k — Preview

This is a preview/browsing repo. The full tracked dataset is at initialneil/TEDUB10k (gated, requires EULA acceptance).

TEDUB10k is the upper-body subset of TEDWB10k — TED speakers cropped to upper-body shots, tracked with a holistic SMPL-X + FLAME + MANO tracker (omni-600 pose, MediaPipe face, WiLoR hands), filtered to frames with a visible face, both hands, and hip.

This repo contains:

Per-subject thumbnails (metadata/previews/<p>/<id>.jpg) — first upper-body source frame
Tracking QC grids (metadata/ehm/<p>/<id>.jpg) — omni-overlay visualizations
Parquet tables with embedded previews for the HF Dataset Viewer

Stats

8758 train / 20 val subjects · ~1,056,182 kept frames
License: CC-BY-NC-ND 4.0 (TED source content)

Full dataset

For frames, mattes, and SMPL-X/FLAME/MANO tracking parameters, visit the gated main repo: initialneil/TEDUB10k