miiidev/rikugan
Local, offline deepfake video detection tool. Upload a video, get a forensic analysis with heatmap overlays, per-region breakdowns, and a downloadable PDF report.
What's novel
Local, offline deepfake video detection tool. Upload a video, get a forensic analysis with heatmap overlays, per-region breakdowns, and a downloadable PDF report.
Code Analysis
10 files read · 3 roundsLocal deepfake video detector that fuses three independent forensic signals (EfficientNet-V2 spatial CNN with Grad-CAM, multi-signal artifact analysis, and rPPG pulse extraction) into a weighted suspicion score with per-region heatmaps and a PDF report.
Strengths
Genuine algorithmic depth across all three detection channels — custom two-level Grad-CAM, four distinct forensic artifact sub-signals with sigmoid scoring, and a proper DSP pipeline for rPPG — all with graceful degradation and transparent, explainable outputs. Clean architecture with protocol-based stores, lazy model loading, cancellation support, and well-structured tests that verify actual algorithmic behavior.
Weaknesses
Test coverage is limited to pure-logic modules (fusion, rPPG, landmarks) with no integration tests for the pipeline, spatial CNN, or artifact detector; scipy is used but not explicitly declared as a dependency. The artifact detector's sigmoid parameters and weights appear hand-tuned without documented justification or ablation.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
55
Contributors
1
Files
231
Active weeks
4
Repository
Language
Python
Stars
1
Forks
0
License
—