Machine Learning
In this first phase of the project, we are focused on recognizing images with a Machine Learning model that will classify images based on the marks. The model is built using TensorFlow, an open-source Machine Learning platform created by Google.
However even though the model gives accurate prediction in the majority of cases (89%), it becomes more tedious to work with when a large number of classes are required. To date, the algorithm has been trained on Niger/Nigeria datasets.
STEP 1: IMAGE SELECTION
The user can start an inference process by uploading an image of a scarification mark and a prediction about the classification according to ethnic group will be displayed (with probability). The Language of Marks model is frequently trained on new drawings and images, which means that the acceptable lists of images will most likely be updated, every time that the model is updated.
LANGUAGE OF MARKS SAMPLE GALLERY
INSTRUCTIONS FOR HOW TO DRAW
Use your mouse to draw a scarification mark / pattern.
- *Do not draw any facial features
- *Draw only one type of scarification pattern at a time
- *Use the "reset" button to clear the drawing
- *If statisfied with your drawing, click "submit" to continue
- *To update your drawing without starting over, add lines and click "submit" again
IMAGE UPLOAD SPECIFICATIONS
* This model is a prototype. For the best results from this model, please review our image guidelines before uploading your image.
- * Image Requirements
- * Image Requirements
- * Image Requirements
STEP TWO: IMAGE CLASSIFICATION
This version of the ML model, built on TensorFlow, an open-source machine learning platform, uses image recognition to classify images based on the scarification patterns contained in the pictures. At the current stage, the model has been trained to classify up to 18 ethnic group categories. Click on "RUN THE MODEL" to see the top 5 ethnonyms the scarification marks in the selected image might represent.
