RekaDaily-10k-raw

# RekaDaily-10k (raw) Raw, unscripted, first-person daily-life video, collected through [Claru](https://claru.ai), Reka's data collection marketplace — recorded by paid collectors in their own homes and workplaces on head-mounted and handheld phones, across multiple regions. **This dataset is being released incrementally and is growing toward the full [10,312-hour release](https://reka.ai/news/rekadaily-10k-egocentric-household-manipulation-data).** Current contents: ~886 hours / 39,643 videos / 1,877 shards. Videos are exactly as collected — no re-encoding, no cuts, no filtering beyond basic integrity checks. A processed tier (short clips with machine captions) is released separately under the same `RekaDaily-10k` prefix. ## Structure Videos are packed into WebDataset tar archives (~5 GB), organized by collection project: ``` data/<project>/shard-NNNNN.tar # pairs of <video_id>.<mp4|mov> + <video_id>.json metadata/browse.parquet # one row per video: 160px thumbnail + every metadata field metadata/index.parquet # the same metadata fields, without thumbnails metadata/thumbs.parquet # full-resolution 320px stills (datasets Image column) sample/ # ~400 loose videos for quick browsing ``` Load the video shards with any WebDataset reader, e.g.: ```python import webdataset as wds url = "https://huggingface.co/datasets/RekaAI/RekaDaily-10k-raw/resolve/main/data/egocentric_household_tasks/shard-00000.tar" ds = wds.WebDataset(url) ``` The Dataset Viewer above opens on the `browse` table: every video as a thumbnail next to its full metadata row, so the collection can be skimmed without downloading anything. The `metadata` config is the same fields without the images, for lighter programmatic reads, and `metadata/thumbs.parquet` keeps the full-resolution 320px stills. The tar shards are not previewed in the viewer (the Hub's WebDataset preview is currently broken platform-wide for video archives); they download and stream normally. Projects: `egocentric_household_tasks`, `egocentric_household_tasks_usa`, `egocentric_commercial_environments`, `residential_egocentric_latam_upload_via_claru`, `video_capture_activities`, `video_capture`, `video_capture_first_person_videos_phone`. ## Metadata fields Each `.json` sidecar carries the fields below; `metadata/index.parquet` and `metadata/browse.parquet` carry the same set, one row per video (`browse` adds the `thumbnail` image column): | field | description | |---|---| | `video_id` | unique id, matches the media file name | | `project` | Claru collection project | | `flow`, `activities` | activity taxonomy (activities-type projects): session scenario + performed actions | | `category`, `subcategory` | category taxonomy (video-capture-type projects) | | `lighting` | lighting condition, where captured | | `duration_s`, `fps`, `width`, `height`, `num_frames`, `codec` | probe stats | | `collector` | salted-hash collector id — distinct values ≈ distinct environments | Each video populates one taxonomy family (flow/activities *or* category/subcategory) depending on its project type. ## Consent, privacy & takedown This dataset was collected through [Claru](https://claru.ai), Reka's data collection marketplace, as described in the [release announcement](https://reka.ai/news/rekadaily-10k-egocentric-household-manipulation-data). Collectors are paid contractors who opt in, and every session is recorded with the wearer's knowledge and agreement. Collectors are instructed to record only with the agreement of other adults present and to keep others out of frame where that is not possible. Every video in this release has been processed to remove container metadata — GPS coordinates, device identifiers, and capture timestamps — and verified clean before upload, in addition to the automated PII screening described in the announcement. Screening is not perfect. If you find something in this release that should not be there, tell us and we will remove it: contact **[e-mail]**. ## License Apache 2.0 — use, redistribute, and build on this data, including commercially, with attribution per the license terms.