awesome-human-activity-recognition

Two-stream adaptive graph convolutional network for skeleton-based action recognition from CVPR 2019.

Unified Python toolkit for time series including classification, clustering, and anomaly detection.

Channel-wise topology refinement graph convolution for skeleton-based action recognition from ICCV 2021.

Reference implementation of the convolutional LSTM architecture for wearable activity recognition.

Basketball activity recognition from a single wrist-worn inertial sensor using deep learning.

Hierarchically decomposed graph convolutional network for skeleton action recognition from AAAI 2024.

Temporal action detection benchmark with 20k untrimmed YouTube videos across 200 classes.

Dense video captioning and temporal grounding with 20k videos and 100k captions.

Unified SMPL motion capture parameters from 40+ datasets covering 16k minutes and 344 subjects.

Spatio-temporal action detection with 430 movie clips and 80 atomic action labels with bounding boxes.

Motion-language alignment dataset with 43 hours and 3.7k sequences annotated with SMPL and text labels.

RGB-D human-object interaction with 3D pose spanning 321 sequences from 20 subjects.

Annual challenge for temporal action detection, proposals, and dense captioning.

Motion-language understanding and temporal action segmentation on mocap data.

CVPR 2025 multi-track benchmark covering ego-pose, action recognition, and language understanding.

Egocentric action recognition, detection, and anticipation competition.

Annual challenge for transportation mode recognition from smartphone sensors.

Human behavior understanding from UAV perspectives with multi-modal data.

Abedin et al., IMWUT 2021, attention mechanisms for multi-sensor HAR.

Tran et al., ICCV 2015, pioneering 3D convolutions for video feature learning.

Ordonez and Roggen, Sensors 2016, establishing deep learning for wearable activity recognition.

Li et al., ACM Computing Surveys 2022, comprehensive review of deep learning approaches for HAR.

Carreira and Zisserman, CVPR 2017, inflating 2D ImageNet architectures to 3D video.

Wang et al., ECCV 2024, scaling video foundation models to 6B parameters across 60+ benchmarks.

Action recognition papers and datasets.

IMU-based sensing for activity recognition and navigation.

Human pose estimation methods and benchmarks.

Self-supervised learning methods applicable to video and sensor modalities.

GCN and transformer methods for skeleton HAR.

Video understanding systems and architectures.

University of Pennsylvania course covering motion representations relevant to HAR.

Interactive textbook chapter on video understanding and action recognition with PyTorch code.

Step-by-step guide to training action recognition models on custom datasets.

Colab notebook for training text-conditioned human motion diffusion models on HumanML3D.

Comprehensive deep learning tutorial for inertial sensor HAR with PyTorch.

Lecture materials covering temporal modeling, two-stream networks, and 3D convolutions for action recognition.

Efficient GPU-accelerated video reader for deep learning training pipelines.

Google's on-device ML framework for pose estimation, hand tracking, and gesture recognition.

Pretrained checkpoints and configs for 100+ action recognition models.

Real-time multi-person keypoint detection for skeleton extraction from video.

Live SOTA tracking across all major HAR benchmarks.

Character animation from video input, useful for activity recognition visualization.

6B-parameter video-language model checkpoints on Hugging Face for action recognition and retrieval.

Pretrained motion encoder transferable to 3D pose estimation, action recognition, and mesh recovery.

Masked video distillation pretrained model competitive with VideoMAE on downstream action recognition.

Efficient video transformer with multi-scale tokens achieving 90.0% top-1 on Kinetics-400.

Billion-parameter video foundation model pretrained on millions of clips, finetunable for action recognition.