Experiment lab
Two ways to probe expression recognition
The teaching simulation uses fixed, published rules to show how conditions change a prediction — no camera, no upload, no model. The photo lab is different: it runs a real pre-trained expression model inside your browser on a photo you choose.
Teaching simulation. All numbers are preset illustrative values produced by a fixed rule. They are not real model accuracy, not calibrated probabilities, and not a judgement about any real person's emotion.
Example class scores
- Smiling-type pattern72.0%
- Neutral-type pattern20.0%
- Frowning-type pattern8.0%
Scores always sum to 100%.
Current output status
Label reported (example output)
'Uncertain' is an output state of the system, not a fourth expression category competing with the others.
Default vs. current conditions
| Class | Default | Current | Change |
|---|---|---|---|
| Smiling-type pattern | 72.0% | 72.0% | 0.0 |
| Neutral-type pattern | 20.0% | 20.0% | 0.0 |
| Frowning-type pattern | 8.0% | 8.0% | 0.0 |
Evidence retained under current conditions: 100%
Which conditions affected this result
All conditions are at their most favourable setting.
Suggested exercise: keep everything else fixed and switch intensity from pronounced to subtle. Then add a mask. Notice that two 'mild' changes together can push the system into the uncertain state.
Terminology: 'subtle' here means a small-amplitude expression in a static image. This is not video-based micro-expression analysis.