|
|
--- |
|
|
license: cc-by-4.0 |
|
|
datasets: |
|
|
- danielritchie/cinematic-mood-palette |
|
|
language: |
|
|
- en |
|
|
tags: |
|
|
- tflite |
|
|
- embedded |
|
|
- emotion |
|
|
- color |
|
|
- hri |
|
|
- robotics |
|
|
- affective-computing |
|
|
- real-time |
|
|
- vad |
|
|
- tiny-model |
|
|
--- |
|
|
|
|
|
# VIBE Color Model |
|
|
|
|
|
A 365-parameter TFLite model that maps emotional state to cinematic color expression. Designed to run on embedded hardware with minimal compute. |
|
|
|
|
|
## Model Description |
|
|
|
|
|
Given a 5-dimensional emotional coordinate (VAD+CC), returns a cinematic visual treatment — not just a color, but RGB plus independent Energy and Intensity parameters drawn from cinematographic practice. |
|
|
|
|
|
**Architecture:** 5→16→12→5 fully connected network |
|
|
**Size:** 3.5KB |
|
|
**Parameters:** 365 |
|
|
**Format:** TFLite (embedded deployment), H5 (inspection/fine-tuning) |
|
|
|
|
|
## Inputs and Outputs |
|
|
|
|
|
**Input:** VAD+CC vector — 5 float values in [0, 1] |
|
|
|
|
|
| Dimension | Meaning | |
|
|
|---|---| |
|
|
| Valence | Negative ↔ Positive emotional tone | |
|
|
| Arousal | Calm ↔ Energized | |
|
|
| Dominance | Passive ↔ Powerful | |
|
|
| Complexity | Minimal ↔ Rich | |
|
|
| Coherence | Chaotic ↔ Harmonious | |
|
|
|
|
|
**Output:** 5 cinematic parameters — 5 float values in [0, 1] |
|
|
|
|
|
| Dimension | Meaning | |
|
|
|---|---| |
|
|
| R | Red channel | |
|
|
| G | Green channel | |
|
|
| B | Blue channel | |
|
|
| Energy | How alive/active the display feels | |
|
|
| Intensity | How pronounced the effect is applied | |
|
|
|
|
|
## Training Data |
|
|
|
|
|
Trained on [danielritchie/cinematic-mood-palette](https://huggingface.co/datasets/danielritchie/cinematic-mood-palette) — ~80 curated anchor points mapping emotional states to visual treatments drawn from film and photography. |
|
|
|
|
|
## Validation |
|
|
|
|
|
Validation is qualitative. The model is evaluated by behavioral coherence — does the output feel cinematically appropriate for the emotional input? Formal quantitative benchmarks are not meaningful for a model of this size and purpose. |
|
|
|
|
|
## Intended Use |
|
|
|
|
|
Part of [VIBE-Eyes](https://github.com/brainwavecollective/vibe-eyes) — a real-time emotional display system for conversational robots. The model runs on-device, receiving VAD+CC vectors from an edge emotion engine and driving LED color output without any cloud dependency. |
|
|
|
|
|
Also useful as a lightweight reference implementation for anyone mapping affective state to visual expression in constrained environments. |
|
|
|
|
|
## Limitations |
|
|
|
|
|
- Small training set (~80 anchor points): functions as a reference structure, not comprehensive coverage |
|
|
- Culturally specific: draws primarily from Western cinematic tradition |
|
|
- Interpretive: mappings reflect observed patterns in film, not objective measurements |
|
|
|
|
|
## License |
|
|
|
|
|
CC-BY-4.0 — use freely with credit |
