How samperochon/smartflat works
Smartflat is a research framework for multimodal healthcare video analysis. Its principal workflow turns recordings and annotations into extracted features, clusters and validates behavioral prototypes, segments time series, assembles symbolic sequences, and supports barycenter and clinical analyses. The graph also preserves the modality-specific extraction branches and annotation/configuration support. The README documents the overall notebook sequence; source excerpts are partial, so internal links not established by those sources are omitted.
Main entry points: build.py Dataset preparation · main.py Recursive prototyping · clustering.py Prototype clustering · loader.py Dataset loading
Components 21
Data preparation
- Dataset preparation
smartflat/datasets/build.py - Clinical annotations
smartflat/annotation_smartflat.py - Dataset loading
smartflat/datasets/loader.py
Feature extraction
- Video features
smartflat/features/video/main.py - Speech features
smartflat/features/audio/main.py - Gaze features
smartflat/features/gaze/main.py - Hand features
smartflat/features/hands/main.py - Pose features
smartflat/features/skeleton/main.py - Hand processing
smartflat/features/hands_processing/main.py
Behavioral analysis
- Prototype clustering
smartflat/engine/clustering.py - Model builders
smartflat/engine/builders.py - Change-point detection
smartflat/engine/change_point_detection.py
Symbolic analysis
- Recursive prototyping
smartflat/features/symbolization/main.py - Symbolic inference
smartflat/features/symbolization/inference.py - Temporal distances
smartflat/engine/distances/_rtwe.py - Barycenter averaging
smartflat/features/symbolic_barycenter/main.py
Study support
- Clinical metrics
smartflat/metrics.py - Analysis configuration
smartflat/configs/loader.py
Other components
- Researcher
- Study recordings
- Study data files
Connections 22
- Researcher → Study recordings: provides
README.md - Study recordings → Dataset preparation: preprocesses
README.md - Dataset preparation → Study data files: reads and writes
- Clinical annotations → Dataset preparation: supports
smartflat/datasets/build.py - Study recordings → Video features: extracts representations
README.md - Study recordings → Speech features: extracts speech
README.md - Study recordings → Gaze features: extracts gaze
README.md - Study recordings → Hand features: extracts landmarks
README.md - Study recordings → Pose features: extracts pose
README.md - Hand features → Hand processing: processes landmarks
- Dataset preparation → Dataset loading: loads prepared data
- Dataset loading → Prototype clustering: supplies datasets
smartflat/engine/clustering.py - Analysis configuration → Prototype clustering: configures
smartflat/engine/clustering.py - Model builders → Prototype clustering: builds components
smartflat/engine/clustering.py - Prototype clustering → Recursive prototyping: prototypes feed
README.md - Dataset preparation → Change-point detection: provides feature sequences
README.md - Change-point detection → Recursive prototyping: segments feed
README.md - Recursive prototyping → Symbolic inference: supports inference
smartflat/features/symbolization/inference.py - Symbolic inference → Barycenter averaging: sequences analyzed
README.md - Temporal distances → Barycenter averaging: computes temporal distance
smartflat/features/symbolic_barycenter/main.py - Recursive prototyping → Clinical metrics: supports group analysis
README.md - Researcher → Clinical metrics: examines results
README.md