Session de Travail 2
We make use of The Recognition Machine to make snapshots with algorithmic classifications of each student.
Classification is based on a technique called "deep learning" — a form of "machine learning" which itself fits into the broader history of Artificial Intelligence. The technique works by using "pre-trained models" : the software is able to quickly automatically label images. These outputs are based on (typically) thousands of examples of "training data" — other images that have been manually labelled with the desired outputs. We use two models, one for age & gender classification, and another for "emotion".
The emotion model is based on the popular FER2013 dataset which has a training set of nearly 30000 images that have been labelled according to the scheme :
- 0 : Angry
- 1 : Disgust
- 2 : Fear
- 3 : Happy
- 4 : Sad
- 5 : Surprise
- 6 : Neutral
Questions are posed about the dataset : Who has collected these images, from where do they come from, who did the labelling ?
Antje shows polaroids from past workshops where participants self-describe themselves reflecting on the practices of colonial photography.
Finally we use The Recognition Machine to match input faces with faces in the archive (via faces from the dataset).
