EgoXtreme: Egocentric Object Pose Data Under Extreme Conditions

Approximately 1.3 million frames of egocentric video, totaling 775.5 minutes (~12.9 hours), were captured at 30 fps using Aria glasses. The dataset is designed for robust 6D object pose estimation under extreme environmental conditions and provides high-resolution 1408 x 1408 raw fisheye RGB images. It was created by taegyoun88 and last updated on the Hugging Face platform in April 2026.

Use Cases

• Training models for 6D object pose estimation based on egocentric fisheye views.
• Benchmarking model robustness based on data captured under extreme environmental conditions.
• Developing algorithms for handling high-resolution 1408 x 1408 fisheye imagery from wearable glasses.

Strengths

• Approximately 1.3 million frames provide a large-scale resource for training.
• High-resolution 1408 x 1408 raw fisheye RGB images offer detailed visual data.
• Captured under extreme conditions, which may support robustness testing.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

taegyoun88 on Hugging Face

Collection Method

Captured at 30 fps using Aria glasses.

Freshness

Last updated 2026-04-06 04:18:00; freshness should be verified.