Scene 02 / 06
Applied computer vision.
See how an AI inference pipeline turns live imagery into face geometry, visible expression labels and hand-gesture signals entirely on-device.
Run on-device AI inference
The browser will ask once. Gestures, raised fingers and visible facial expressions are described on-device; video is never uploaded.
IDLE / LOCAL
Pixels are
data.
Computer vision becomes useful data science when models can extract reliable geometric information beyond curated training imagery.
My TUM master’s thesis evaluated how CNN, U-Net and DeepLabV3 models generalize in image-based façade semantic segmentation.
01
Represent
Transform imagery into measurable features and semantic classes.
02
Generalize
Evaluate model performance across façades, viewpoints and conditions.
03
Operate
Connect model inference to geospatial and infrastructure data workflows.