Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LSC-ADL: An Activity of Daily Living (ADL)-Annotated Lifelog Dataset Generated via Semi-Automatic Clustering

Minh-Quan Ho-Le, Duy-Khang Ho, Van-Tu Ninh, Cathal Gurrin, Minh-Triet Tran

TL;DR

The paper addresses the gap in lifelog retrieval by integrating Activities of Daily Living (ADLs) as a semantic, temporally-aware layer. It introduces LSC-ADL, an ADL-annotated lifelog subset derived from LSC22, produced via a semi-automatic labeling pipeline that combines LLaMA-3.2-vision captions, GPT-driven label proposals, and human-in-the-loop verification, followed by HDBSCAN-based clustering to capture intra-class variation. The dataset comprises 35 ADL classes and reveals a long-tailed distribution with meaningful daily and diurnal activity patterns, highlighting potential gains in retrieval explainability and context-aware search. Overall, LSC-ADL provides a structured resource to advance lifelog retrieval, egocentric action recognition, and interpretable lifelog analytics, with downloadable annotations to support reproducibility and broad adoption.

Abstract

Lifelogging involves continuously capturing personal data through wearable cameras, providing an egocentric view of daily activities. Lifelog retrieval aims to search and retrieve relevant moments from this data, yet existing methods largely overlook activity-level annotations, which capture temporal relationships and enrich semantic understanding. In this work, we introduce LSC-ADL, an ADL-annotated lifelog dataset derived from the LSC dataset, incorporating Activities of Daily Living (ADLs) as a structured semantic layer. Using a semi-automatic approach featuring the HDBSCAN algorithm for intra-class clustering and human-in-the-loop verification, we generate accurate ADL annotations to enhance retrieval explainability. By integrating action recognition into lifelog retrieval, LSC-ADL bridges a critical gap in existing research, offering a more context-aware representation of daily life. We believe this dataset will advance research in lifelog retrieval, activity recognition, and egocentric vision, ultimately improving the accuracy and interpretability of retrieved content. The ADL annotations can be downloaded at https://bit.ly/lsc-adl-annotations.

LSC-ADL: An Activity of Daily Living (ADL)-Annotated Lifelog Dataset Generated via Semi-Automatic Clustering

TL;DR

The paper addresses the gap in lifelog retrieval by integrating Activities of Daily Living (ADLs) as a semantic, temporally-aware layer. It introduces LSC-ADL, an ADL-annotated lifelog subset derived from LSC22, produced via a semi-automatic labeling pipeline that combines LLaMA-3.2-vision captions, GPT-driven label proposals, and human-in-the-loop verification, followed by HDBSCAN-based clustering to capture intra-class variation. The dataset comprises 35 ADL classes and reveals a long-tailed distribution with meaningful daily and diurnal activity patterns, highlighting potential gains in retrieval explainability and context-aware search. Overall, LSC-ADL provides a structured resource to advance lifelog retrieval, egocentric action recognition, and interpretable lifelog analytics, with downloadable annotations to support reproducibility and broad adoption.

Abstract

Lifelogging involves continuously capturing personal data through wearable cameras, providing an egocentric view of daily activities. Lifelog retrieval aims to search and retrieve relevant moments from this data, yet existing methods largely overlook activity-level annotations, which capture temporal relationships and enrich semantic understanding. In this work, we introduce LSC-ADL, an ADL-annotated lifelog dataset derived from the LSC dataset, incorporating Activities of Daily Living (ADLs) as a structured semantic layer. Using a semi-automatic approach featuring the HDBSCAN algorithm for intra-class clustering and human-in-the-loop verification, we generate accurate ADL annotations to enhance retrieval explainability. By integrating action recognition into lifelog retrieval, LSC-ADL bridges a critical gap in existing research, offering a more context-aware representation of daily life. We believe this dataset will advance research in lifelog retrieval, activity recognition, and egocentric vision, ultimately improving the accuracy and interpretability of retrieved content. The ADL annotations can be downloaded at https://bit.ly/lsc-adl-annotations.

Paper Structure

This paper contains 12 sections, 4 figures.

Table of Contents

  1. Introduction
  2. Related work
  3. Action recognition
  4. Action recognition on egocentric lifelog data
  5. Methodology
  6. Original dataset
  7. Label selection
  8. Label proposal pipeline
  9. Exploratory data analysis
  10. Conclusion
  11. Acknowledgements
  12. Labels and Descriptions

Figures (4)

  • Figure 1: Label Selection Pipeline – The process begins with a random sampling of 20 days x 15 temporally ordered images from the LSC dataset. These images are processed by the LLaMA-3.2-11b-vision model to generate captions, which are concatenated into higher-level descriptions. Simultaneously, raw ADL labels are extracted from other ADL-related works. Both the generated descriptions and raw labels are fed into a GPT model for new label generation and categorization before further manual processing into final ADL classes.
  • Figure 2: Label Proposal Pipeline – The process begins with initializing the golden corpus using Decompse-Expand prompting technique houenhancing and SnapSeek retrieval system snapseek, followed by generating label suggestions. Annotators review the suggestions and approve or modify them before re-clustering and updating the corpus iteratively.
  • Figure 3: Label distribution across the dataset.
  • Figure 4: Hour distribution per label.