VidupeDotNet

VidupeDotNet is a modern, high-performance video deduplication utility built on .NET 10 and Blazor Server. It identifies duplicate or near-duplicate video files using 7 comparison algorithms — CLIP, LPIPS, SSIM, pHash, dHash, Histogram, and MSE — with optional GPU acceleration.

📦 Docker image available on Docker Hub: hub.docker.com/r/gafda/vidupe-net

---

Features

• 7 Comparison Algorithms: CLIP (semantic), LPIPS (deep perceptual), SSIM, pHash, dHash, Histogram, MSE — ordered by accuracy. Custom mode combines any algorithms. GPU-recommended warning for neural-network algorithms.
• GPU Acceleration: Optional CUDA/OpenCL acceleration with automatic CPU fallback.
• Selective Folder Scanning: Catalog all folders but scan/compare only those you check. Uncheck folders to exclude them without removing them.
• Collapsible Folder Tree: Folder paths are presented as a path-compressed tree with tri-state parent checkboxes, tree connector lines, and aggregate video counts.
• Dark & Light Themes: Toggle between dark and light modes with persistent preference.
• High Performance: Multi-core async processing for fast scanning of large libraries.
• Deep Scanning: Configurable frame capture positions (1×1 through 4×4 grids, cut-ends mode, others) for thorough visual comparison.
• Mirrored Video Detection: Detects horizontally flipped duplicates (e.g. front-camera recordings).
• Video Summary Pre-filter: Optional majority-vote pHash per video for O(1) pair rejection, reducing comparison time and RAM for large collections.
• Lossless WebP Thumbnails: Captured frames are stored as lossless WebP for 100% faithful pixel reproduction.
• Retry Failed Scans: One-click retry for all files that failed during scanning.
• Re-compare Warnings: Alerts when comparison data may be stale after a new scan, with optional suppression.
• Pair Estimate: Live estimate of comparison pair count and thumbnail size in the sidebar.
• Incomplete Thumbnail Rescan: Videos with missing captures are automatically re-scanned.
• Persistent State: Scan data, ignored pairs, and failed scan history are preserved across sessions.
• Docker Ready: Pre-built images for both x64 and ARM64 (Raspberry Pi 4/5) with optional GPU passthrough.
• Connection Resilience: Automatic reconnection with user notification when the server disconnects.

---

Getting Started with Docker

The recommended way to run VidupeDotNet is via the pre-built Docker image on Docker Hub.

Prerequisites

• Docker or Podman
• FFmpeg is bundled in the image — no separate installation needed
• Optional: NVIDIA GPU + drivers on the host for GPU-accelerated algorithms (pHash, SSIM)
• Optional: CUDA Toolkit 12.x + cuDNN 9.x on the host for neural-network GPU acceleration (CLIP, LPIPS) — see GPU Setup

Quick Start

docker run -d \
  --name vidupe-net \
  -p 8080:8080 \
  -v /path/to/your/videos:/videos:ro \
  -v vidupe-data:/data \
  gafda/vidupe-net:latest

Then open your browser at http://localhost:8080.

⚠️ Read-only volume warning: The :ro flag mounts your video folder as read-only. This prevents VidupeDotNet from deleting or renaming any files on disk. If you want to allow in-app deletion or renaming of video files, remove the :ro flag:

-v /path/to/your/videos:/videos

Only do this if you intend to allow the application to modify your files directly.

Available Tags

Tag	Architecture	Description
latest	x64	Latest stable release
latest-arm64	ARM64	Latest stable release for ARM64 devices
1.7.0	x64	Specific version
1.7.0-arm64	ARM64	Raspberry Pi 4/5 and other ARM64 devices

---

Volume Configuration

Volume	Container Path	Purpose
Your video library	/videos	The folder VidupeDotNet will scan
Persistent data	/data	Database, cache, and app state

You can mount multiple video subfolders if needed:

docker run -d \
  --name vidupe-net \
  -p 8080:8080 \
  -v /mnt/nas/movies:/videos/movies:ro \
  -v /mnt/nas/series:/videos/series:ro \
  -v vidupe-data:/data \
  gafda/vidupe-net:latest

⚠️ Read-only volume warning: All volumes above use :ro (read-only). File deletion and renaming from within the app will be blocked. Remove :ro on any mount where you want to allow modifications.

---

GPU Setup

VidupeDotNet has two independent GPU acceleration paths. Both are optional — the app always falls back to CPU gracefully.

Mode	Algorithms	Host Requirement
CUDA/OpenCL	pHash, SSIM	NVIDIA/AMD GPU + drivers
ONNX CUDA	CLIP, LPIPS	NVIDIA GPU + CUDA Toolkit 12.x + cuDNN 9.x

The NVIDIA driver alone is not sufficient for ONNX CUDA acceleration. You must also install the CUDA Toolkit libraries on the host machine:

# Fedora / RHEL / Bazzite
sudo dnf install cuda-cudart-12-6 libcublas-12-6 libcufft-12-6 libcurand-12-6 libcusparse-12-6 libcudnn9-cuda-12

# Ubuntu / Debian
sudo apt install cuda-toolkit-12-6 libcudnn9-cuda-12

If CUDA libraries are missing, CLIP and LPIPS will fall back to CPU automatically with a warning in the logs.

---

GPU Passthrough

The container image is GPU-agnostic — GPU drivers are injected at runtime by the host.

GPU Vendor	Docker flag	Podman flag
NVIDIA	--gpus all	--device nvidia.com/gpu=all (CDI)
AMD	--device /dev/kfd --device /dev/dri	--device /dev/kfd --device /dev/dri
Intel	--device /dev/dri	--device /dev/dri

NVIDIA (Docker)

docker run -d \
  --name vidupe-net \
  -p 8080:8080 \
  --gpus all \
  -v /path/to/your/videos:/videos:ro \
  -v vidupe-data:/data \
  gafda/vidupe-net:latest

NVIDIA (Podman, CDI)

podman run -d \
  --name vidupe-net \
  -p 8080:8080 \
  --device nvidia.com/gpu=all \
  -v /path/to/your/videos:/videos:ro \
  -v vidupe-data:/data \
  gafda/vidupe-net:latest

AMD (Docker or Podman)

docker run -d \
  --name vidupe-net \
  -p 8080:8080 \
  --device /dev/kfd --device /dev/dri \
  -v /path/to/your/videos:/videos:ro \
  -v vidupe-data:/data \
  gafda/vidupe-net:latest

Intel (Docker or Podman)

docker run -d \
  --name vidupe-net \
  -p 8080:8080 \
  --device /dev/dri \
  -v /path/to/your/videos:/videos:ro \
  -v vidupe-data:/data \
  gafda/vidupe-net:latest

⚠️ Read-only volume warning: All examples above use :ro on the video mount. To enable in-app file deletion or renaming, remove the :ro suffix from the -v flag for your video folder.

---

Docker Compose

A minimal docker-compose.yml for CPU-only usage:

services:
  vidupe-net:
    image: gafda/vidupe-net:latest
    container_name: vidupe-net
    ports:
      - "8080:8080"
    volumes:
      - /path/to/your/videos:/videos:ro   # Remove :ro to allow file deletion/renaming
      - vidupe-data:/data
    restart: unless-stopped

volumes:
  vidupe-data:

For GPU passthrough, add the appropriate device or deploy section for your vendor. See the GPU Passthrough section above.

---

ARM64 (Raspberry Pi 4/5)

Pull and run the ARM64 image directly:

docker run -d \
  --name vidupe-net \
  -p 8080:8080 \
  -v /path/to/your/videos:/videos:ro \
  -v vidupe-data:/data \
  gafda/vidupe-net:1.7.0-arm64

Note: Neural-network algorithms (CLIP, LPIPS) will run on CPU on ARM64 as ONNX CUDA is not supported on this architecture.

---

How it Works

1. Discover — Recursively scan video files across your selected folders.
2. Extract Metadata — Read duration, resolution, codec, and bitrate from each file.
3. Capture Frames — Sample frames at configurable positions throughout each video.
4. Compare — Run the selected algorithms to score visual similarity between video pairs.
5. Review — Side-by-side comparison UI with one-click delete, swap, or ignore.

---

Origin & Credits

VidupeDotNet is an independent implementation inspired by the original vidupe created by Kristian Koskimaki.

While this is a completely separate codebase, it honors the legacy of the original tool by continuing its mission to provide a powerful, open-source solution for video library management.

---

License

Distributed under the GPL v3.0 License. See LICENSE for details.