👉 Chinese version for BIT user / 中文

AutoSlides Extractor

Automatically extract slide images from video presentations using advanced computer vision

Overview • Features • Download • Quick Start • How It Works • Build

---

Overview

AutoSlides Extractor is a powerful cross-platform desktop application designed to intelligently identify and extract distinct slides from presentation videos. Using sophisticated computer vision algorithms—specifically Structural Similarity Index (SSIM) and Perceptual Hashing (pHash)—it automatically detects when slide content changes significantly and saves each unique slide as a high-quality image.

New in v1.1.0: AI-powered slide classification using a MobileNetV4 ONNX model to automatically identify and remove non-slide content (desktop screens, black screens), a trash review system, and a PDF Maker for exporting slides.

✨ Key Features

Core Capabilities	Advanced Tech
🎯 Intelligent Slide Detection Two-stage algorithm using SSIM for precise slide extraction	⚡ Hardware Acceleration GPU support via CUDA, OpenCL, Metal, DirectX, and Vulkan
🧹 Smart Post-Processing Auto-removal of redundant slides using pHash & exclusion lists	🤖 AI-Powered Classification MobileNetV4 model filters out non-slide content
🗑️ Trash Review System Safely review and restore removed slides with metadata tracking	🧩 High Performance Multi-threading & SIMD optimizations (AVX2, NEON)
📄 PDF Maker Organize and export extracted slides to compressed PDF documents	🎼 Multi-Format Support MP4, AVI, MOV, MKV, WMV, FLV, WebM

Detailed Features

• Configurable Sensitivity: Presets (Strict, Normal, Loose) and custom SSIM thresholds.
• Batch Processing: Queue multiple videos for sequential processing.
• Memory Optimization: Chunk-based processing for handling large video files efficiently.
• Cross-Platform: Native look and feel on macOS, Windows, and Linux.

📥 Download & Installation

Pre-built Releases

Download the latest installer or portable package for your operating system:

Platform	Download	Note
macOS	AutoSlides.Extractor-macOS-arm64.dmg	Apple Silicon only. Run quarantine fix command below.
Windows	Setup.exe / Portable.zip	64-bit Windows 10 or later.

Important

macOS Users: After installation, run the following command in Terminal to allow the app to run:

sudo xattr -d com.apple.quarantine /Applications/AutoSlides\ Extractor.app

System Requirements

• OS: macOS 11+ (Apple Silicon), Windows 10+ (64-bit)
• CPU: Quad-core processor recommended
• RAM: 8 GB minimum (16 GB for large videos)
• GPU: Recommended for hardware acceleration

🚀 Quick Start

1. Launch AutoSlides Extractor.
2. Add Videos: Drag & drop video files or click "Add Videos".
3. Configure (Optional): Select an Output Directory and adjust SSIM sensitivity if needed.
4. Start: Click the Start button.
5. Review: Once finished, check the output folder. Use the Trash Review to recover any mistakenly removed slides or the PDF Maker to compile them.

🛠️ Usage Guide

1. Main Workflow

• Input: Supports major video formats. Videos are queued and processed one by one.
• Monitoring: Real-time status updates show:
  • 🟡 Queued: Waiting.
  • 🔵 Processing: FFmpeg decoding -> SSIM analysis -> ML Classification.
  • 🟢 Completed: Done with slide count.
• Controls: Pause, Resume, or Remove items from queue easily.

2. Post-Processing & AI

• Redundancy Removal: Uses pHash to find and delete near-duplicate images.
• Exclusion Lists: Define patterns (like Intro/Outro slides) to automatically ignore.
• AI Classification (v1.1.0):
  • "slide": Kept.
  • "not_slide": Removed (e.g., desktop, black screen).
  • "may_be_slide": Configurable action (keep or delete).
  • Tip: Use the visual range sliders in Settings to tune confidence thresholds.

3. PDF Export

• Navigate to PDF Maker.
• Select the root folder containing your extracted slides.
• Choose sort order (Name/Date) and output quality (resize/compress).
• Generate a single PDF document for your presentation.

🎯 How It Works

1. Stage 1: Change Detection (SSIM)
   • Samples frames (I-frames) and calculates structural similarity.
   • Significant drops in similarity signal a potential new slide.
2. Stage 2: Stability Verification
   • Checks subsequent frames to ensure the "new slide" is stable and not just a transition effect.
3. Stage 3: Deduplication (pHash)
   • Computes perceptual hashes of extracted images.
   • Removes duplicates even if minor pixel noise exists.
4. Stage 4: AI Filtering (MobileNetV4)
   • Classifies image content to remove non-slide elements like desktop wallpapers or empty screens.

⚙️ Technical Details

Architecture: Video Input → Hardware Decoder → SSIM Analysis → Image Output → pHash Deduplication → ML Classification → PDF Export

Performance:

• SIMD: SSE4.2, AVX2, AVX512, NEON.
• GPU Accel: CUDA, DirectML, Metal, OpenCL.
• Inference: Core ML (macOS), CUDA/DirectML (Windows).

🔧 Building from Source

Prerequisites:

• C++17 Compiler, CMake 3.16+
• Qt 6 (Core, Widgets, Gui)
• OpenCV 4.x, FFmpeg

Build Steps:

git clone https://github.com/bit-admin/AutoSlides-Extractor.git
cd AutoSlides-Extractor
mkdir build && cd build

# Configure
cmake ..
# Or with specific optimizations
# cmake -DENABLE_CUDA=ON ..

# Build
cmake --build . --config Release

❓ Troubleshooting

Issue	Solution
"App is damaged" on macOS	Run sudo xattr -d com.apple.quarantine /Applications/AutoSlides\ Extractor.app in Terminal.
GPU not used	Ensure you have the latest drivers installed. On Windows, check if CUDA is installed for NVIDIA cards.
Slides missed	Try the "Strict" SSIM preset or decrease the custom threshold (e.g., to 0.995).
Too many duplicates	Enable Post-Processing and lower the Hamming Threshold (e.g., to 8).

📄 License

This project is licensed under the MIT License.

---

Made with ❤️ using Qt6, OpenCV, and FFmpeg