Awesome Human Activity Recognition

Human Activity Recognition (HAR) is the field of recognizing human actions and activities from sensor data — including video, skeleton/mocap, wearable IMU, and multimodal egocentric inputs. This list covers 53 datasets, frameworks, pretrained models, tutorials, papers, competitions, and tools for HAR research.

 

---

Quick Stats

Modality	Datasets	Highlights
Vision (RGB/Depth)	14	Kinetics-700, UCF-101, ActivityNet, AVA
Skeleton & MoCap	7	NTU RGB+D 60/120, AMASS, Human3.6M
Wearable Sensors	13	UCI-HAR, PAMAP2, CAPTURE-24 (3883 hrs)
Multimodal & Egocentric	7	Ego4D (3.3k hrs), EPIC-Kitchens-100
Emerging & Frontier	12	HumanML3D, Motion-X++, Ego-Exo4D

---

Repository Architecture

graph LR
    subgraph Datasets["53 Datasets"]
        V["Vision (14)"]
        S["Skeleton (7)"]
        W["Wearable (13)"]
        M["Multimodal (7)"]
        E["Emerging (12)"]
    end

    subgraph Ecosystem
        F["Frameworks & Libraries"]
        P["Pretrained Models"]
        T["Tutorials & Courses"]
    end

    subgraph Automation
        LC["Link Check (weekly)"]
        SU["SOTA Update (weekly)"]
    end

    Datasets --> F
    Datasets --> P
    F --> T
    SU -->|updates| Datasets
    LC -->|validates| Datasets

---

Which Dataset Should I Use?

Pick your modality and task, then follow the recommendation.

Video ClassificationTemporal Action DetectionSkeleton / MoCapWearable SensorsEgocentric / MultimodalText-to-Motion Generation

Start with Kinetics-700 for pretraining, evaluate on UCF-101 or HMDB-51 for comparison with prior work.

ActivityNet for proposals, AVA for spatio-temporal, MultiTHUMOS for dense multi-label.

NTU RGB+D 120 is the de facto standard. For text-motion alignment, use BABEL or HumanML3D.

UCI-HAR for baselines, PAMAP2 for multi-sensor, CAPTURE-24 for real-world scale (151 subjects, 3883 hours).

Ego4D for scale (3.3k hours), EPIC-Kitchens-100 for kitchen actions, Ego-Exo4D for cross-view (CVPR 2024).

HumanML3D for single-person, InterHuman for two-person, Motion-X++ for whole-body with face and hands.

---

Explore

• Datasets — Browse all 53 dataset cards organized by modality
• Taxonomy — Multi-dimensional classification by task, license, scale, and year
• Surveys — Curated survey papers across all modalities
• Benchmarking — SOTA baselines and performance bands per dataset
• Roadmap — What is coming next
• Contributing — How to add datasets or improve the list

---

Citation

@misc{awesome_har_2025,
  title   = {Awesome Human Activity Recognition: A Curated List},
  author  = {Wenxuan Huang},
  year    = {2025},
  url     = {https://github.com/Leooo-Huang/awesome-human-activity-recognition},
  note    = {GitHub repository}
}