Scene Detection Policies and Keyframe Extraction Strategies for Large-Scale Video Analysis

2025 paper comparing scene-detection policies and keyframe extraction strategies at scale.

Link

Related resources

  • LMSKE — Keyframe Extraction for Video SummarizationLarge-model sequential keyframe extraction pipeline combining TransNetV2 shot segmentation, CLIP features, and adaptive clustering.
  • video-black-frame-detection — black/scene boundary detectorFFmpeg + OpenCV black-frame detector for scene/ad-break/chapter boundaries, with optional LLM analysis and JSON/SRT output via REST API or…
  • Katnapip-installable MIT library for automated keyframe extraction (LUV color-diff), video compression, and smart image cropping.
  • MKV-Chapter-Maker — auto chapters via black/silence/sceneAutomatically generates MKV chapters using black-frame, silence or scene detection and writes them via mkvpropedit without re-encoding.
  • VidChapters-7M (Vid2Seq)NeurIPS 2023 dataset-and-benchmark project plus Vid2Seq model for automatic video chapter generation at scale.
  • SceneSeg (Local-to-Global Scene Segmentation)CVPR 2020 multimodal movie scene segmentation using place and audio features on the MovieNet dataset; reference ML baseline with code.
  • AutoShotCVPR-NAS 2023 shot boundary detection model with a NAS-optimized architecture and the new SHOT dataset (853 videos, 11,606 annotations).
  • butteraugli (Rust port)Pure Rust port of Butteraugli from libjxl with XYB color, visual masking, and multi-scale bands, validated to within 0.001% of the referenc…
  • SimpleVQADeep-learning no-reference video quality assessment for user-generated content, in PyTorch with SlowFast feature extraction.
  • AutoSubSyncSubtitle auto-sync GUI wrapping ffsubsync/autosubsync/alass; can sync against a reference subtitle instead of the video.