awesome-human-activity-recognition
Always up-to-date, most comprehensive HAR resource — continuously scanned and auto-updated from Papers with Code. 53 datasets integrated across all modalities.
What is awesome-human-activity-recognition?
Always up-to-date, most comprehensive HAR resource — continuously scanned and auto-updated from Papers with Code. 53 datasets integrated across all modalities.
Overview of awesome-human-activity-recognition
What it is
This is an always up-to-date, comprehensive resource for Human Activity Recognition (HAR). It is continuously scanned and auto-updated from Papers with Code, integrating 53 datasets across all modalities including video, skeleton/mocap, wearable IMU, and multimodal egocentric inputs. The list covers datasets, frameworks, pretrained models, tutorials, papers, competitions, and tools for HAR research.
Key features
- Continuously scanned and auto-updated from Papers with Code - 53 datasets integrated across all modalities (vision, skeleton, wearable, multimodal, emerging)
- Covers datasets, frameworks, pretrained models, tutorials, papers, competitions, and tools - Includes a “Which Dataset Should I Use” guide for modality and task selection - Multilingual README (Chinese, German, Spanish, French, Japanese, Korean, Portuguese, Russian)
- Organized by modality: Vision, Skeleton, Wearable, Multimodal, Emerging
Use cases
- Video action classification using Kinetics-700, UCF-101, HMDB-51 - Temporal action detection in untrimmed video using ActivityNet, AVA, MultiTHUMOS - Skeleton-based action recognition using NTU RGB+D 120 - Wearable sensor activity recognition using UCI-HAR, PAMAP2, CAPTURE-24 - Egocentric and multimodal understanding using Ego4D, EPIC-Kitchens-100 - Text-to-motion generation using HumanML3D, InterHuman
Getting started
See the repository README for details.
License and pricing
Licensed under CC-BY-4.0. Open source.
Based on the project's public README and repository metadata.