# Recorded public examples

These four validation clips are from **Auslan-Daily: Australian Sign Language
Translation for Daily Communication and News**, by Xin Shen, Shaozu Yuan,
Hongwei Sheng, Heming Du and Xin Yu (NeurIPS 2023). They are served as a small
randomly selected catalogue in the research walkthrough.

- [Dataset website and download terms](https://uq-cvlab.github.io/Auslan-Daily-Dataset/docs/en/dataset-download)
- [Dataset paper](https://proceedings.neurips.cc/paper_files/paper/2023/hash/feb34ce77fc8b94c85d12e608b23ce67-Abstract-Datasets_and_Benchmarks.html)
- [Creative Commons Attribution 4.0 International](https://creativecommons.org/licenses/by/4.0/)

The current dataset download page states and links **CC BY 4.0** (verified 25
September 2026). The 2023 paper's footnote 8 instead names CC BY-NC-ND 4.0;
that discrepancy is recorded rather than hidden. This is a noncommercial
academic research showcase. No endorsement by the dataset authors, programme
creators or performers is implied. The licence gives no warranties and other
rights may apply, as described in the linked licence.

Each source is a signer-only clip from the official validation split. The display
copy is silent H.264 for browser compatibility, with no additional crop. The
frame count and 25 fps time grid are preserved. The poster is the first source
frame. DWPose tracking, normalised coordinates, encoder activations,
self-attention and the English prediction were produced by this project.
Overlays are drawn separately in the browser. Each clip's manifest records its
source hash, display hash, processing and model provenance.

The page chooses from this fixed, attributed catalogue in the browser. It does
not expose the rclone OAuth configuration, mount the 17 GB source archive, or
accept arbitrary remote paths. This keeps the public endpoint bounded and
prevents a dataset-wide file browser. The examples are selected validation
illustrations, not representative performance or live inference.