---
title: Aegis-Safe-Work Fall Detector
colorFrom: yellow
colorTo: red
sdk: gradio
sdk_version: 5.23.3
app_file: app.py
pinned: false
license: cc-by-nc-nd-4.0
---

# Aegis-Safe-Work: Fall Detector

Fall detection over short video clips using EfficientNet-Lite0 combined with
a temporal attention mechanism (Attention MLP) and a binary classifier. This
Space loads the trained checkpoint from
[`beaunix/aegis-fall-detector`](https://huggingface.co/beaunix/aegis-fall-detector)
and runs it on ZeroGPU.

## How it works

1. The uploaded video (max 45 seconds) is opened and 16 frames are sampled
   uniformly across its full duration (not a sliding window).
2. Each frame is letterboxed (aspect-ratio preserved, black padding) to
   224x224 and normalized with ImageNet statistics, matching the training
   ETL exactly.
3. All 16 frames are processed in a single GPU forward pass: EfficientNet-Lite0
   extracts per-frame features, temporal attention weights and pools them,
   and the MLP classifier outputs a single fall probability for the clip.
4. The report shows the per-frame attention weights as a bar chart, the
   frame with peak attention overlaid with the verdict, and a metrics
   summary table.

## Model performance (validation set)

| Metric    | Value  |
|-----------|--------|
| Accuracy  | 0.9762 |
| F1        | 0.9730 |
| Precision | 0.9574 |
| Recall    | 0.9890 |
| Threshold | 0.65   |

## Notes

- Videos longer than 45 seconds are rejected to protect ZeroGPU quota.
- This Space demonstrates the fall classifier in isolation, independent of
  the other Safe-Work models (PPE detection, fire/smoke detection).

## License

CC BY-NC-ND 4.0 (Attribution - NonCommercial - NoDerivatives).
See https://creativecommons.org/licenses/by-nc-nd/4.0/