Michael Chan, an AI developer, recently published a medium article showing how to build a facial expression recognition algorithm to detect the emotion of world leaders when talking about the Corona Virus.
Micheal has used HaarCascades for face detection and the following CNN architecture for recognizing emotion:
- Input layer of size (48x48x1): grayscale
- 4 Convolutions of filters (3×3) or (5×5) each of them followed by a Max-Pool (2×2)
- 2 Full connected Layers of size 256 and 512 (in this order)
- Output layer of size (7×1) which is for the 7 labels of emotions
The CNN model was trained on a Kaggle dataset.
Although the Facial Expression Recognition algorithm is on point, it must be understood that this is something that is just for entertainment and it is not right to say that the world leaders are lying just because they are feeling ‘nervous’ or ‘sad’.