EgoVista Pilot V0.2: Egocentric Annotations for Daily-Life Manipulation Activities

EgoVista Pilot v0.2 is an annotations-only dataset for first-person human activity analysis, generated by the EgoVista pipeline. It contains per-frame annotations including 2D pose and hand keypoints, 3D hand keypoints, 16-bit depth maps, segmentation masks, contact flags, and action labels. The dataset was created by author Leodocr and last updated on Hugging Face in May 2026.

Use Cases

• Training models for 2D and 3D hand pose estimation based on the provided keypoint annotations.
• Developing segmentation models for hands and objects using the included RLE COCO masks.
• Researching human-object contact and manipulation dynamics based on the per-frame contact flags.
• Benchmarking depth estimation models using the 16-bit depth maps generated by Depth Anything V2.
• Classifying or recognizing daily-life manipulation activities based on the provided action labels.

Strengths

• Provides a multi-modal annotation set including pose, depth, segmentation, and action labels for a unified analysis of egocentric scenes.
• Annotations were generated by the EgoVista pipeline, suggesting a degree of automated consistency.
• Includes both 2D and 3D hand keypoints, which may support more advanced spatial reasoning tasks.

Limitations

• Description metadata is limited; actual data quality, annotation accuracy, and consistency require manual inspection after download.
• Row count, file formats, and license are unknown, which may limit suitability assessment and legal use.
• The dataset is annotations-only; the original RGB video frames are not included, limiting its use to tasks that can rely solely on derived annotations.

Provenance

Source

Leodocr on Hugging Face.

Collection Method

Generated by the EgoVista pipeline.

Time Range

null

Freshness

Last updated 2026-05-02 11:02:00; freshness should be verified.

Geography

null