How petuuu/sign-language-recognizer works
This is a Finnish/ASL static sign-letter recognition project. Its user-facing entry point selects image-to-landmark export, annotated image detection, or live video detection; a separate model-training script reads exported landmarks and trains an MLP. The README describes recognition, but the sampled interactive path does not connect detected landmarks to model inference, and the model code shows training rather than an end-user prediction flow. The graph reflects those evidenced paths and omits unshown runtime links.
Main entry points: image_pipeline.py CSV export · train.py Training loop · main.py Mode selection · helpers.py Dataset split
Components 16
Detection workflows
- Mode selection
src/main.py - CSV export
src/image_pipeline.py - Image detection
src/image_pipeline.py - Video detection
src/image_pipeline.py
Landmark processing
- Hand detector
- Landmark normalization
src/image_pipeline.py - Landmark annotation
src/image_pipeline.py
Model training
- Landmark CSV
- Dataset split
src/model/helpers.py - Training loop
src/model/train.py - MLP architecture
src/model/architecture.py - Model helpers
src/model/helpers.py - Training plots
src/model/train.py
Other components
- User
- Image files
- Camera
Connections 20
- User → Mode selection: chooses mode
src/main.py - Mode selection → CSV export: runs export
src/main.py - Mode selection → Image detection: runs detection
src/main.py - Mode selection → Video detection: runs detection
src/main.py - CSV export → Image files: reads images
src/image_pipeline.py - Image detection → Image files: reads images
src/image_pipeline.py - Video detection → Camera: captures frames
src/image_pipeline.py - CSV export → Hand detector: detects hands
src/image_pipeline.py - Image detection → Hand detector: detects hands
src/image_pipeline.py - Video detection → Hand detector: detects hands
src/image_pipeline.py - CSV export → Landmark normalization: normalizes points
src/image_pipeline.py - CSV export → Landmark CSV: writes landmarks
src/image_pipeline.py - Image detection → Landmark annotation: draws landmarks
src/image_pipeline.py - Video detection → Landmark annotation: draws landmarks
src/image_pipeline.py - Training loop → Dataset split: loads and splits
src/model/train.py - Dataset split → Landmark CSV: reads rows
src/model/helpers.py - Training loop → MLP architecture: trains model
src/model/train.py - Training loop → Model helpers: uses loss and metrics
src/model/train.py - MLP architecture → Model helpers: uses activations
src/model/architecture.py - Training loop → Training plots: saves metrics
src/model/train.py