Again, if we have decided to satisfy only the minimal requirements for using the face model and the Visual Recognition service, we can get away with collecting only 10 images for each class which we intend our model to train on (10 happy faces, 10 sad faces, and 10 negative faces).
These individual training files will need to be the following:
- Either JPEG (.jpg) and PNG (.png) formatted
- At least 32*32 pixels in size
- Compressed as a group into class ZIP files, that is, 10 happy faces in a happy.zip file and 10 sad faces in a sad.zip file
A sampling of our initial happy model training data is shown in the following screenshot:
The preceding images are of 10 faces showing what we think can be labeled as being representative of happy facial expressions. Notice that the individual files have all been added to a compressed (ZIP) file named happy.zip.
A sampling of our initial sad model training data is shown in the following screenshot:
And clearly, as the former group displayed happiness, the later images are of faces showing what we think can be labeled as being representative of sad expressions (individual files and the zipped file, sad.zip).