Original article was published on Deep Learning on Medium
ML Assisted Image Tagging
Image Classification Multi-Class, Image Classification Multi-Label, Object Identification (Bounding Box)
To Apply Machine learning or Deep Learning on any image or vision based project first images has to be tagged. Tagging image is labor intensive work and take long time. How can we make it much more productive is what we are going to see.
To solve the above we are going to use the Azure Machine learning service — Data Labelling features which has Manual and ML Assited tagging.
What is ML Assisted Tagging?
Lets take like 2000 pictures and take two tags or 1000 images with one tag for example like face mask. For the below tutorial i am using only few images but for ML assit to work we need more images. Once you collected the image then tag around 100 of them for 1000 images manual. Options availble to have multiple people can tag the images.
Once the images are tag then wait until the ML assit does it job. For ML assit we need GPU based cluster and also make sure select a region where GPU based virtual machines are available.
Steps to do
- Collect images
- Create a Blob Storage
- Create a container
- Upload the images
- Create a Azure Machine learning Services — i choose WEST US 2 since GPU was available
- Create a Data Labeling project
- Create Data Set from Data Store
- Create project
- Image Tagging
- ML Assit tagging
Collect images for the below sample, download images from pubic domain search engines.
Create a Blob Storage
Create a container
Upload the images
Create a Azure Machine learning Services — i choose WEST US 2 since GPU was available Create a Data Labeling project
Create Data Set from Data Store
Add Class or Label
Select ML Assit
Select the project
Click Data Label
ML Assit tagging
Tag all the images until the UI says it says there is no tasks
Go back to Project Screen
Wait until remaining is all completed.
In our say if you had 1000 images you need to tag 100 and then it takes time to do ML assisted tagging.