Datasets:
Tasks:
Robotics
Formats:
csv
Languages:
English
Size:
1K - 10K
ArXiv:
Tags:
egocentric-vision
first-person-video
embodied-ai
robot-learning
video-language
vision-language-action
License:
Maintenance
How to keep the atlas accurate, consistent, and reviewable. The catalog is machine-checked, so most mistakes are caught by the validator before review.
Add a resource
- Confirm it is in scope. Human or animal first-person capture, or ego-exo
where the ego view is central. Robot-only / multi-view-robotic / driving /
general long-video belong in Adjacent and Related Resources (
scope: adjacent). - Find an official source. Project page, GitHub, Hugging Face, arXiv, OpenReview, university page, Zenodo, or an official challenge page. No blog reposts.
- Add the YAML entry in
data/resources.ymlusing resource_schema.md. Reuse task/modality tokens from taxonomy.md; add new tokens todata/taxonomy.yml. - Add the README row in the matching section, keeping the table format.
- Set
verified_atto today. - Run
ruby scripts/build_artifacts.rbto refresh generated counts, dates, CSV/JSON, and figure labels.
Verify access status
- Apply status_policy.md. Prefer the conservative status.
- Do not infer licenses — if you cannot confirm one, write
unknown(in the access matrix) and leavelicenseoff the YAML entry. - Record what is missing in
release_noteforwatch/partialentries.
Update the README
- New resource → add a table row.
- Changed status/scale → edit the row and the YAML in the same PR.
- New section/figure → add it to the Contents list.
- Badge counts, the
Updated:date, the At-a-Glance row, and fallback site stats are generated byruby scripts/build_artifacts.rb.
Update the YAML
- Keep
nameidentical to the README when possible. If the display label must differ, document it indata/readme_aliases.yml. - Names must be unique; check the duplicate guard if the validator complains.
Update badge and date
- The headline
resources-<N>badge counts egocentric resources (scope is notadjacent). - Bump
meta.last_major_auditindata/resources.yml, then runruby scripts/build_artifacts.rb. Do not edit the generated README date or badge manually.
Run local checks
ruby scripts/validate_catalog.rb
ruby scripts/build_artifacts.rb --check
ruby scripts/build_readme_i18n.rb --check
ruby scripts/build_hf_package.rb --check
ruby scripts/audit_catalog.rb
Fix every error. Warnings are non-fatal but should be resolved when practical:
missing verification dates, missing release_note on watch/partial entries,
missing license fields on open datasets, stale access checks, or taxonomy gaps.
Build the Hugging Face package
ruby scripts/build_hf_package.rb /private/tmp/awesome-egocentric-hf
The package copies publishable files only and prepends
huggingface/dataset_card_header.txt
to the temporary README. The GitHub README must stay metadata-free.
PR checklist
- Resource is in scope (egocentric, or correctly marked
scope: adjacent). - Official source linked;
statusis conservative. - Scale numbers match the official source (not inflated).
- Both
README.mdanddata/resources.ymlupdated;namematches. - New task/modality tokens added to
data/taxonomy.yml. -
meta.last_major_auditupdated if needed, then generated artifacts rebuilt. -
ruby scripts/validate_catalog.rbandruby scripts/build_artifacts.rb --checkpass with no errors.