Gaming DatasetsDefine a useful pilot
A collection brief should specify what success looks like in a small sample before it sets a large hours target. An action-conditioned world model may need ordered action–observation sequences; a behavior-cloning policy may need consistent mappings from observations to player actions. Neither is defined by video resolution alone.
This is a planning template, not a dataset download or a promise that every requested game, signal or format is available. The fields below describe your requested scope. They do not change recorder settings.
What to agree before collection
- Scope: title versions, tasks, player variation, expected session length and the unit used to count accepted hours.
- Synchronization: timebase, action interval convention, encoding behavior and how interruptions are represented.
- Quality: reject conditions, pilot review, reporting for missing frames and any restrictions on menus or overlays.
- Evaluation: splits by player, session or task where those identifiers exist. Do not randomly distribute neighboring frames across training and evaluation.
- Delivery: schema version, checksums, file naming, permissions and the process for resolving failed checks.
Use the dataset evaluation guide to review a sample, and the video–action alignment guide to define timing requirements.
Share it with your team
The blank JSON template is available without registration. You may copy, adapt and share the collection template for your own projects; attribution to Gaming Datasets is appreciated but optional. Downloaded custom briefs include a source link so collaborators can find the field explanations.
When you are ready, contact us about a collection project and describe the pilot you want to evaluate.