🎬 Segmenter

Segmenter is a high-performance visual timestamp annotation tool for Linux and Windows. It lets you create, edit, and upload media segment markers — Intro, Recap, Credits, and Preview — to TheIntroDB and IntroDB.

---

✨ Features

🎧 Audio Waveform Analysis

Automatically extracts audio and displays a density waveform on the timeline. Uses FFT-based spectral flatness to color-code passages — mint for dialogue, orange for music-heavy themes like intros and outros.

🖼️ Frame Strip Preview

Real-time MJPEG-based frame strip centered on the playhead for frame-accurate seeking and positioning.

🤖 GPU-Accelerated Season Fingerprinting

Integrated Recurring Content Detector (RCD) scans an entire season directory and automatically finds duplicate segments across episodes using:

• FAISS-GPU nearest-neighbor search (NVIDIA) or FAISS-CPU (AMD / no GPU)
• CNN feature extraction (MobileNetV3 or SimpleCNN fallback via PyTorch)
• Color Histogram and Color Texture Moment modes for CPU-only setups
• AMD ROCm support — PyTorch ROCm builds accelerate CNN feature extraction on AMD GPUs
• NVIDIA CUDA support — full GPU acceleration including FAISS-GPU

Results are cached in /tmp and reused automatically on subsequent loads.

🌐 Remote Network Shares

Stream and analyze files directly from Samba/SMB, SFTP, NFS, WebDAV, or HTTP mounts via native file dialogs.

🔑 Local Configuration

API keys are stored in ~/.config/Segmenter/keys.json — no keyring daemon required.

---

⌨️ Controls & Shortcuts

Timeline Mouse Actions

Action	Description
Click	Seek to position
Alt + Drag	Create a new segment
Shift + Drag	Move a segment between rows
Drag Edges	Resize segment boundaries
Scroll Wheel	Zoom in/out around playhead

Keyboard Shortcuts

Shortcut	Description
Space	Toggle play / pause
← / →	Step one frame backward / forward
I / Shift+I	Set Intro start / end
R / Shift+R	Set Recap start / end
C / Shift+C	Set Credits start / end
P / Shift+P	Set Preview start / end
,	Snap nearest segment edge to playhead

---

🚀 Quick Start (Linux)

Prerequisites

Dependency	Purpose
python3 (≥ 3.10)	Runtime
pip	Package manager
ffmpeg	Audio/video processing
NVIDIA CUDA drivers	Optional — full GPU acceleration (CNN + FAISS)
AMD ROCm + PyTorch-ROCm	Optional — GPU acceleration for CNN feature extraction

Run

./linux/run.sh

This script automatically creates a virtual environment, installs all Python dependencies, and launches the app.

---

🚀 Quick Start (Windows)

Prerequisites

Dependency	Purpose
.NET 8.0 SDK	Compilation and runtime
ffmpeg	Audio extraction and thumbnail generation (must be in system PATH)

Build & Run

Run the PowerShell build script:

.\windows\build.ps1

This will restore dependencies, compile the project, and generate a standalone executable at windows\dist\Segmenter.exe.

---

📦 Packaging

Build standalone .deb, .rpm, and Arch Linux packages:

./linux/build_packages.sh

Output artifacts are placed in linux/dist/.

---

🏗️ Project Structure

.
├── windows/
│   ├── build.ps1           # Windows Build Script
│   ├── Segmenter/          # WPF C# Application
│   │   ├── MainWindow.xaml # Main UI
│   │   └── ...
│   └── dist/               # Compiled Segmenter.exe
├── linux/
│   ├── app.py              # Entry point
│   ├── ui.py               # Main window & all UI logic
│   ├── timeline.py         # Interactive timeline widget
│   ├── framestrip.py       # MJPEG frame strip
│   ├── player.py           # Video player (mpv backend)
│   ├── audio.py            # Audio waveform extraction
│   ├── models.py           # Data models (SegmentType, SegmentDraft, etc.)
│   ├── clients.py          # TheIntroDB & IntroDB API clients
│   ├── parser.py           # Filename → TMDB metadata parser
│   ├── validator.py        # Segment validation rules
│   ├── rcd_integration.py  # RCD scan dialog & worker
│   ├── gpu.py              # Unified GPU detection (CUDA / ROCm / CPU)
│   ├── rcd/                # Recurring Content Detector engine
│   │   ├── detector.py     # Main detection pipeline
│   │   ├── featurevectors.py # Feature vector extraction (CH, CTM, CNN)
│   │   ├── video_functions.py # Video resize & framerate utils
│   │   └── evaluation.py   # Detection evaluation metrics
│   ├── run.sh              # Launcher script
│   ├── build_packages.sh   # Packaging script
│   └── requirements.txt    # Python dependencies
├── docs/
│   └── screenshot.png
├── LICENSE
└── README.md

---

🙏 Acknowledgements

• recurring-content-detector — Core fingerprinting, FAISS search, and CNN algorithms used for automated recurring segment detection.
• IntroStamp — The original macOS segment editor that served as UX model and design inspiration.
• TheIntroDB & IntroDB — Public segment marker databases and APIs.
• TMDB — Movie and TV metadata.

---

📄 License

MIT License — see LICENSE.