How harshavardhannayaka20072007-png/activity-14-hackathon-preparation works
PixelGuard is an image-forensics prototype with two distinct capabilities: interactive analysis of one uploaded image, and an offline CASIA-compatible dataset feature pipeline. In the app, an upload is validated before the user starts analysis; ELA, image statistics, and EXIF are then displayed. The dataset pipeline discovers and inspects labeled images, extracts forensic features, groups duplicates for leakage-safe splits, and scales features using training data. Classifier, localization, evaluation, and explainability modules are Phase 2 placeholders, not working prediction stages. The graph reflects sampled calls and README claims; unsampled internals are not inferred.
Main entry points: app.py Streamlit app · dataset_pipeline.py Pipeline orchestration · features.py Forensic features · ela.py ELA generation
Components 15
Interactive analysis
- Streamlit app
app/app.py
Forensic analysis
- Image validation
src/preprocessing.py - ELA generation
src/ela.py - Basic features
src/features.py - Forensic features
src/features.py - EXIF extraction
src/metadata.py
Dataset pipeline
- Pipeline orchestration
src/dataset_pipeline.py - Dataset discovery
src/dataset.py - Dataset inspection
src/dataset.py - Feature and split files
src/dataset_pipeline.py
Planned capabilities
- Classification placeholder
src/classifier.py - Region localization
src/localization.py - Model evaluation
src/evaluation.py - Prediction explanations
src/explainability.py
Other components
- Image uploader
Connections 11
- Image uploader → Streamlit app: uploads image
app/app.py - Streamlit app → Image validation: validates image
app/app.py - Streamlit app → ELA generation: requests ELA
app/app.py - Streamlit app → Basic features: extracts statistics
app/app.py - Streamlit app → EXIF extraction: extracts EXIF
app/app.py - Streamlit app → Image uploader: displays analysis
app/app.py - Forensic features → ELA generation: uses ELA
src/features.py - Pipeline orchestration → Dataset discovery: loads records
src/dataset_pipeline.py - Pipeline orchestration → Forensic features: extracts features
src/dataset_pipeline.py - Pipeline orchestration → Dataset inspection: inspects dataset
README.md - Pipeline orchestration → Feature and split files: writes tables
src/dataset_pipeline.py