- Notifications
You must be signed in to change notification settings - Fork1
ai-forever/dynamic_gestures
Folders and files
| Name | Name | Last commit message | Last commit date | |
|---|---|---|---|---|
Repository files navigation
This repository contains using HaGRID dataset for dynamic gesture recognition. The dataset is availablehere.
├── ocsort/ # source code for Observation-Centric Sorting│ ├── kalmanfilter.py # Kalman filter│ ├── kalmanboxtracker.py # Kalman box tracker│ ├── association.py # Association of boxes with trackers├── utils/ # useful utils│ ├── action_controller.py # Action controller for dynamic gestures│ ├── box_utils_numpy.py # Box utils for numpy│ ├── enums.py # Enums for dynamic gestures and actions│ ├── hand.py # Hand class for dynamic gestures recognition│ ├── drawer.py # Debug drawer├── onnx_models.py # ONNX models for gesture recognition├── main_controller.py # Main controller for dynamic gestures recognition, uses ONNX models, ocsort and utils├── run_demo.py # Demo script for dynamic gestures recognitionClone and install required python packages:
git clone https://github.com/ai-forever/dynamic_gestures.git# or mirror link:cd dynamic_gestures# Create virtual env by conda or venvconda create -n dynamic_gestures python=3.9 -yconda activate dynamic_gestures# Install requirementspip install -r requirements.txt
To run demo, you just need to runrun_demo.py script.
python run_demo.py --detector<path_to_detector> --classifier<path_to_classifier> --debug
--detector (optional) Path to the hand detector model.Default:models/hand_detector.onnx
--classifier (optional) Path to the crops classifier model.Default:models/crops_classifier.onnx
--debug (optional) Enables debug mode to see bounding boxes and class labels.
Next, we will show dynamic gestures in user mode and debug mode. In user mode, we show only the final result of dynamic gesture recognition. In debug mode, we show the result of each step of dynamic gesture recognition:
- hand detection
- hand tracking
- gesture recognition
- action recognition
At the moment the code supports 6 groups of dynamic gestures:
     |      |
This work is licensed under a variant ofApache License, Version 2.0.
Please see the specificlicense.
You can cite the paper using the following BibTeX entry:
@misc{nuzhdin2024hagridv21mimagesstatic, title={HaGRIDv2: 1M Images for Static and Dynamic Hand Gesture Recognition}, author={Anton Nuzhdin and Alexander Nagaev and Alexander Sautin and Alexander Kapitanov and Karina Kvanchiani}, year={2024}, eprint={2412.01508}, archivePrefix={arXiv}, primaryClass={cs.CV}, url={https://arxiv.org/abs/2412.01508},}