A set of scripts to leverage a dataset of faced-mask for Covid crisis for Watson Visual Recognition
Solutions have been made to detect people wearing masks like in France XXII. My goal is to create a model from Open Source Data using Watson Visual Recognition, a Watson AI services dedicated to do image classification and custom object detection. I wanted to show how easy it is to have a COD model using Watson technologies.
One of the eky challenge when you want to do a visual recognition model is to have a good dataset. Several exists on Internet. One colleague has published one on Github, I have used another one publicly available on Kaggle. That dataset is labelled using YOLO format.
I have written several scripts to implement the different steps of the overall process.
transform_label.js to adapt the content in YOLO to Watson format
upload.js to upload images and data to Watson VR instance
list.js to list uploaded images
train.js to launch and monitor the training process
The scripts has to be run at the root of the extracted Kaggle dataset.
Change the api key, collectionId (in some files) and gateway url (if relevant).
const visualRecognition = new VisualRecognitionV4({
version: '2019-02-11',
authenticator: new IamAuthenticator({
apikey: '**XXXXXXXXXXXXXXXX**'
}),
url: 'https://api.us-south.visual-recognition.watson.cloud.ibm.com',
});
In progress