Skip to content
GrekoLabs.
Back to portfolio ↗

COMPUTER VISION / INTERACTIVE CPU DEMO

Emotion AI

From an image to an estimated expression. One detected face, a visible crop, and CPU inference using an existing trained Keras model.

  • Python
  • TensorFlow
  • Keras
  • OpenCV
  • Next.js

Try Emotion AI

Select an image with one frontal face. JPG or PNG, up to 8 MB. The image is processed in server memory and is not saved.

What the prototype does

  1. Select a JPG or PNG image.
  2. Requires exactly one detected face and displays its crop.
  3. Preprocesses the crop as grayscale, 48 × 48, nearest and float32 / 255.
  4. Displays happy, neutral, sad or surprised, with raw outputs in the saved class order.

If no face or multiple faces are detected, select another image. The model loads without training or changing its weights.

Interpretable results, visible limitations

Scores are model outputs, not certainty percentages or measurements of a person's actual emotional state. Frontal detection can fail on profiles, occlusions or poor lighting. Technical tests do not measure classifier accuracy.