Yoga pose estimation, using Computer Vision, Deep learning and Tensorflow running on Red Hat OpenShift Kubernetes
Grant Steinfeld [email protected] @gsteinfeld https://www.linkedin.com/in/grant-steinfeld/
[Tweet post event](- https://twitter.com/gsteinfeld/status/1235952319944175616?s=20)
https://www.slideshare.net/GrantSteinfeld/cv-deep-learning-on-ibm-redhat-openshift
https://www.meetup.com/ibmcodenyc/events/268172307/
-
Deploy MAX image classifier to RedHat OpenShift 3 on IBM Cloud
-
Build a web application that recognizes Yoga poses using a model from the Model Asset Exchange.
Yogait is a yoga assistant that uses the Human Pose Estimator MAX Model to guess which yoga pose a user is performing. The MAX model is deployed to a local server which performs inference and returns the pose guess to the client application.
- Collin Abidi [email protected]
- Daniel Jalova [email protected]
Pattern Github Repo
MAX Model Github Repo
In this code pattern, we will create an interactive web application that uses the webcam to recognize the user's yoga poses.
- IBM Model Asset Exchange: A place for developers to find and use free and open source deep learning models.
- Docker: Docker is a tool designed to make it easier to create, deploy, and run applications by using containers.
- Python: Python is a programming language that lets you work more quickly and integrate your systems more effectively.
- JQuery: jQuery is a cross-platform JavaScript library designed to simplify the client-side scripting of HTML.
- Bootstrap 3: Bootstrap is a free and open-source front-end library for designing websites and web applications.
The Human Pose Estimator model detects humans and their poses in a given image. The model first detects the humans in the input image and then identifies the body parts, including nose, neck, eyes, shoulders, elbows, wrists, hips, knees, and ankles. Next, each pair of associated body parts is connected by a "pose line"; for example, as the following image shows, a line may connect the left eye to the nose, while another may connect the nose to the neck.
Each pose line is represented by a list [x1, y1, x2, y2], where the first pair of coordinates (x1, y1) is the start point of the line for one body part, while the second pair of coordinates (x2, y2) is the end point of the line for the other associated body part. The pose lines are assembled into full body poses for each of the humans detected in the image.
The model is based on the TF implementation of OpenPose model. The code in this repository deploys the model as a web service in a Docker container. This repository was developed as part of the IBM Developer Model Asset Exchange.
Yogait uses a pre-trained SVM to classify poses. Instead of using the Cartesian lines that the MAX model returns, Yogait uses a Polar representation to perform classification. This was done to make it much easier to classify poses. Instead of training the SVM on a x-y coordinate system, which would require translation and rotation when augmenting data, the polar representation relies only upon the location of the joints relative to the center of the estimated model.
The [x,y] coordinates are converted to [phi, rho] for each joint.
The SVM performs classification on a flattened version of the polar vectors. Compared to a Cartesian representation, this polar representation uses very little data and can perform classification on a human in any part of a captured frame. If the Cartesian representation was to be used, then the user would have to perform all poses in the center of the camera frame.
When the reader has completed the Code Pattern, they will understand how to:
- Build a Docker image of the Human Pose Estimator MAX Model
- Deploy a deep learning model with a REST endpoint
- Generate a pose estimation for a person in a frame of video using the MAX Model's REST API
- Run a web application that using the model's REST API
- Server sends the captured video frame-by-frame from the webcam to the Model API.
- Web UI requests the pose lines estimated for the frame from the Server.
- Server receives data from the Model API and updates the result to the Web UI.
Yogait can be run either in the Jupyter notebook (demo.ipynb) or by running the scripts. You can run Yogait with either a pre-trained model (located at /assets/classifier.pkl) or retrain it on your own data.
NOTE: The set of instructions in this section are a modified version of the one found on the Human Pose Estimator Project Page
To run the docker image, which automatically starts the model serving API, run:
docker run -e CORS_ENABLE=true -e WERKZEUG_RUN_MAIN=true -it -p 5000:5000 codait/max-human-pose-estimator
This will pull a pre-built image from Docker Hub (or use an existing image if already cached locally) and run it. If you'd rather build the model locally you can follow the steps in the model README.
Note that currently this docker image is CPU only (we will add support for GPU images later).
The API server automatically generates an interactive Swagger documentation page.
Go to http://localhost:5000 to load it. From there you see the API with the test requests.
Use the model/predict endpoint to load a test file and get estimated pose lines for the image from the API.
The samples folder contains images you can use to test out the API, or you can use your own.
You can also test it on the command line, for example:
curl -F "file=@path/to/pose.jpg" -X POST http://localhost:5000/model/predict
{
"status": "ok",
"predictions": [
{
"human_id": 0,
"pose_lines": [
{
"line": [
444,
269,
392,
269
]
},
...
{
"line": [
439,
195,
464,
197
]
}
],
"body_parts": [
{
"part_id": 0,
"part_name": "Nose",
"score": "0.83899",
"x": 428,
"y": 205
},
...
{
"part_id": 17,
"part_name": "LEar",
"score": "0.81776",
"x": 464,
"y": 197
}
]
},
...
]
}
Clone the Yogait repository locally by running the following command:
git clone https://github.com/dzwietering/yogaitThen change directory into the local repository
cd yogaitThe optional recommendation for Python development is to use a virtual environment (venv). To install and initialize a virtual environment, use the venv module on Python 3 (you install the virtualenv library for Python 2.7). You can also skip this and continue to install into your current Python environment directly.
# Create the virtual environment using Python. Use one of the two commands depending on your Python version.
# Note, it may be named python3 on your system.
$ python -m venv mytestenv # Python 3.X
$ virtualenv mytestenv # Python 2.X
# Now source the virtual environment. Use one of the two commands depending on your OS.
$ source mytestenv/bin/activate # Mac or Linux
$ ./mytestenv/Scripts/activate # Windows PowerShellBefore running this web app you must install its dependencies:
pip install -r requirements.txtTIP 💡 To terminate the virtual environment use the
deactivatecommand.
The python scripts are located in the main and utils folder.
helpers.py and data.py contain helper functions for the rest of the code
train.py retrains the SVM on the folders and images located in the assets/images folder
inference.py runs a live inference demo that uses the deployed Docker image to estimate live poses
If you don't have the local Docker image running you can edit line 7 of utils/helpers.py to refer to the web API similar to the edit in main.jsas specified in the previous section.
You then start the live inference demo by running:
python inference.py
You can then view live pose estimation in a window and predictions from the SVM in the terminal.
The Human Pose Estimator endpoint must be available as specified in utils/helpers.py for the web app to successfully start.
You can retrain the SVM on your own poses by adding images and folders to assets/images. Images of the same class must be in the same folder and must be .png, .jpeg, or .jpg filetypes.
You can then retrain the SVM on the updated classes by running from the main folder:
```python
python train.py
```
- Model Asset eXchange (MAX)
- Center for Open-Source Data & AI Technologies (CODAIT)
- MAX Announcement Blog
Create / login to your free IBM Cloud Account https://ibm.biz/BdqDVL
- Artificial Intelligence Code Patterns: Enjoyed this Code Pattern? Check out our other Artificial Intelligence Code Patterns
- AI and Data Code Pattern Playlist: Bookmark our playlist with all of our Code Pattern videos
- Watson Studio: Master the art of data science with IBM's Watson Studio
- Deep Learning with Watson Studio: Design and deploy deep learning models using neural networks, easily scale to hundreds of training runs. Learn more at Deep Learning with Watson Studio.
This code pattern is licensed under the Apache Software License, Version 2. Separate third party code objects invoked within this code pattern are licensed by their respective providers pursuant to their own separate licenses. Contributions are subject to the Developer Certificate of Origin, Version 1.1 (DCO) and the Apache Software License, Version 2.