Flikforge Sample Dataset

EgoCentric-Household-Tasks

~100 rights-cleared egocentric household-task clips with dense, fully customizable action annotations.

Build Your Dataset License Your Content

A sample dataset of ~100 rights-cleared household-task video clips with dense per-clip action annotations, drawn from the Flikforge catalog. Each clip carries its source MP4 plus a machine-generated ActionSummary 2.0 JSON sidecar produced by Flikforge's Bootstrapped Dense Supervision (BDS) labeling pipeline. The annotation schema is fully customizable: the ActionSummary 2.0 sidecars shown here are one configuration, and Flikforge engineers bespoke schemas per delivery — tell us the fields, granularity, and taxonomy your model needs and the same footage ships labeled to your specification. Clips span 7 test task categories included as samples: Cleaning Counters (17), Miscellaneous (17), Organization (17), Folding Laundry (16), Dish Washing (15), Vacuuming (13), and Object Movement (8).

Part of the Flikforge EgoCentric collection. Captured on iPhone and GoPro. Egocentric human-demonstration data (passive; no robot action stream). Every clip is rights-cleared with chain-of-title provenance documented via Flikforge's patent-pending AI Rights Certificates.

Dataset structure

Each task has a top-level folder (Dish_Washing/, Folding_Laundry/, Vacuuming/, Cleaning_Counters/, Organization/, Object_movement/, Miscellaneous/) holding one <asset_id>.mp4 and one <asset_id>_ActionSummary2.0.json per clip. A machine-readable catalog.parquet at the repo root indexes every clip with its task, relative media paths, per-clip duration/FPS/resolution, annotation source pipeline, and Flikforge parameter and package identifiers.

Catalog

The 7 task categories below are test categories included as samples; production deliveries are scoped to the task taxonomy your program requires.

task	clips	annotation pipeline(s)
Cleaning Counters	17	Robotics-Cleaning-Training-v1.0, Test-Collection-Indexing-v1.0
Miscellaneous	17	Test-Collection-Indexing-v1.0, Scientific-Sampling-Analysis-v1.0
Organization	17	Test-Collection-Indexing-v1.0, Test-Procedure-Indexing-v1.0
Folding Laundry	16	Folding-Shots-Shirts-Socks
Dish Washing	15	Robotics-Dishwashing-Embodiment-v1.0
Vacuuming	13	Robotics-Vacuum-Training-v1.0
Object Movement	8	Test-Procedure-Indexing-v1.0

Per-clip duration, FPS, and resolution are recorded in catalog.parquet.

Loading

from pathlib import Path
import json
import pandas as pd
from huggingface_hub import snapshot_download

local = Path(snapshot_download("Flikforge/EgoCentric-Household-Tasks", repo_type="dataset"))
catalog = pd.read_parquet(local / "catalog.parquet")

row = catalog.iloc[0]
video_path = local / row["video_path"]                 # decode with decord, torchvision, or PyAV
annotation = json.loads((local / row["annotation_path"]).read_text())  # ActionSummary 2.0

Videos are standard H.264 MP4 and decode with any common video reader. Annotations follow the Flikforge ActionSummary 2.0 schema; see docs/ for the full field contract.

Documentation

Full delivery docs are in docs/: DELIVERY_README.md (the complete delivery contract, including the ActionSummary 2.0 field reference) and DATA_CARD.md (content, provenance, labeling methodology, and limitations).

Terms

Provided by Flikforge, Inc. (flikforge.com). See LICENSE. This sample set is provided for evaluation purposes. For commercial licensing, catalog-scale volumes, and custom annotation schemas, contact info@flikforge.com or visit flikforge.com/datasets. AI Rights Certificate documentation is available to licensees.

Citation

@misc{flikforge_egocentric_2026,
  title  = {EgoCentric-Household-Tasks},
  author = {Flikforge, Inc.},
  howpublished = {\url{https://flikforge.com}},
  year   = {2026},
  note   = {Rights-cleared household-task video sample set with dense, customizable action annotations}
}

Two Paths Forward

Ready to Engineer Training Data for Your Model?

Work with Flikforge →