Egoit 99K: Multimodal Egocentric Life-Logging with 99,000 Records

99,000 multimodal records across audio, image, text, and tabular formats were released by lmms-lab in March 2025 for egocentric life-logging research. The dataset is associated with the EgoLife project and the arXiv:2503.03803 publication.

Use Cases

• Training multimodal models using the `image` and `audio` modalities for egocentric activity recognition
• Developing cross-modal alignment algorithms between `text` descriptions and `image` frames
• Analyzing behavioral patterns using the `tabular` metadata associated with daily life-logging events

Strengths

• 99,000 records
• Four modalities: audio, image, text, and tabular
• Parquet format for efficient data handling

Limitations

• Missing explicit column definitions in the provided metadata
• Geographic bias toward the United States as indicated by region tags

Provenance

Source

lmms-lab (arXiv:2503.03803)

Freshness

Last updated March 2025.

Geography

United States