github Breakthrough/DVR-Scan v2.0-release
DVR-Scan v2.0
on GitHub

47 minutes ago

Release Notes

DVR-Scan v2.0 is a major release that includes an improved new user experience, overhauls timestamp handling for VFR, adds settings presets, and brings drag-and-drop support. Under the hood, motion events are now tracked using presentation timestamps (PTS), which means variable framerate (VFR) videos are handled correctly with default parameters. This release has a huge amount of bugfixes and features, and also improves the online documentation.

Changelog

  • [breaking] The default output mode is now encode (MP4/H.264 via ffmpeg) instead of opencv (.avi), so motion events now include audio and are saved as .mp4 by default. If ffmpeg is not installed and no output mode is set, DVR-Scan falls back to opencv with a warning. Set output-mode = opencv in a config file (or use -m opencv) to restore the previous behavior.
  • [breaking] Remove the -roi/--region-of-interest option and the region-of-interest config option (deprecated since v1.6). Use the region editor (-r/--region-editor), -a/--add-region, or -R/--load-region instead.
  • [breaking] Remove --use-pts option; presentation timestamps are now always used for event timing (the use-pts config option is deprecated and has no effect)
  • [feature] New Scan Wizard provides a simplified step-by-step workflow (add videos, pick a preset, choose output, scan) #263
    • New installs start in the wizard; existing installs keep the full settings view (now called Advanced Mode). You can switch between modes at any time
    • Set the default with the new File -> Startup Mode menu or the startup-mode config option (wizard or advanced), or force a mode for a single launch with dvr-scan-app --startup-mode wizard|advanced
    • Installs without a user config file (e.g. pip) start in the wizard unless DVR-Scan has run before (only when launched without -c/--config or --ignore-user-config); the choice is saved to a new user config file. Downgrading to v1.x afterwards requires removing startup-mode from dvr-scan.cfg, as older versions reject unknown options
    • Start/end time and detection regions set in Advanced Mode also apply to wizard scans and are listed on the summary step
    • Errors when starting a scan (e.g. an unreadable video) are shown in a dialog instead of failing silently, and output modes that need ffmpeg are marked unavailable when it is not installed
  • [feature] Settings presets: a new preset dropdown in the app allows applying built-in presets or saving your own #263
    • User presets are config files stored in a presets folder next to the user config file (Settings -> Open Presets Folder)
    • The Default preset is the user config file itself, replacing the Save As User Default and Reset To User Default menu items; saving it preserves config options the app has no setting for (e.g. startup-mode, verbosity)
  • [feature] Video files can now be dragged and dropped onto the app window to add them as inputs; dropping a .cfg file loads it as the active settings #263
  • [feature] Add select all with Ctrl+A (Command+A on Mac), multiple selection with Shift/Ctrl+click, and remove selection #269
  • [feature] Add a --max-events N flag to allow stopping after detecting N motion events #261
  • [feature] New -Y/--max-event-length option and max-event-length config option: events longer than the given length are split into consecutive events of at most that length (no post-event padding is added at a split; if motion continues, the next event starts where the previous one ended) #271
  • [feature] New --merge-window/merge-window option controls how close two groups of motion must be for them to be merged into the same event; -tb/-tp are now pure padding added around each event and no longer affect merging. The default (auto) uses the same value as time-post-event, which matches the merging behavior of previous versions. Setting time-post-event larger than merge-window caps the post-event padding to the merge window with a warning. #195 #72
  • [feature] Motion mask output (-mo/--mask-output) now encodes high-quality MP4 (H.264) videos via ffmpeg by default when ffmpeg is available, respecting encode-args #275
  • [feature] The Region Editor now supports seeking across the input video timeline via a frame slider and ,/. frame-stepping shortcuts, and starts from the configured start-time in the GUI #248
  • [feature] Add timecode-list-format config option to control the format of the event list printed at the end of a scan (and output in -q/--quiet mode): parts (start-end,start-end,..., syntax of mkvmerge --split parts:), timecodes (default, start,end,start,end,..., syntax of mkvmerge --split timecodes:), or lines (one start-end pair per line) #78
  • [general] Reported event timecodes are now exact rather than quantized to whole frames; formatted timecodes may differ by up to 1ms from previous versions. Frame numbers shown in reports and overlays are approximations on VFR inputs.
  • [general] When using --frame-skip, the post-event window now covers the exact amount of time specified rather than rounding down to a multiple of the skip interval, and back-to-back events share their boundary timecode instead of leaving a one-frame gap.
  • [feature] Frames are now decoded with PyAV by default (input-mode = pyav), which provides more accurate timing and stream metadata; set input-mode = opencv to restore the previous behavior
  • [feature] Multiple input videos are concatenated on a continuous PTS-based timeline with support for seeking across inputs; frame rates no longer need to match exactly
  • [general] Videos with a delayed start (nonzero stream start time) now report event timecodes relative to the start of the file with all input modes
  • [general] Reported event boundaries may shift by a frame or two compared to previous releases, as PyAV and OpenCV decode frames with slightly different color conversion
  • [feature] New --report option writes detected events to a CSV or JSON file (format set by extension), including in scan-only mode; the GUI's Save Report dialog can also save JSON reports. CSV reports now include exact start/end times in seconds (frame numbers are approximations on variable framerate inputs). #221
  • [feature] New output mode encode produces MP4 (H.264) files by piping decoded frames to ffmpeg, combining the quality of ffmpeg mode with support for overlays (-bb/-tc/-fm), multiple inputs, and combined output (-o, video-only); audio from the source video(s) is included, including for events spanning multiple inputs (on VFR inputs, see the note on playback speed below). Encoder settings can be customized with the new encode-args config option. #197
  • [bugfix] Variable framerate (VFR) videos are now handled correctly: event start/end times are always derived from presentation timestamps #20
  • [bugfix] Motion events extracted from multiple input videos are now named after the source video containing the start of each event, instead of always using the first input's filename #258
  • [bugfix] Thumbnails (--thumbnails highscore) are now named after the source video containing each event when scanning multiple inputs, and are numbered correctly in scan-only mode (previously each event's thumbnail overwrote the last one)
  • [bugfix] Using --thumbnails highscore together with -o/--output no longer
    overwrites the combined output video with the thumbnail image; thumbnails are named after the combined output file instead
  • [bugfix] Fixed audio misalignment in encode mode for inputs with a nonzero container start time (e.g. DVR .ts recordings)
  • [general] encode mode output uses the yuv420p pixel format (H.264 High profile) for compatibility with most players unless -pix_fmt is set in encode-args, and odd frame sizes (e.g. mask output with -df 3) are padded to even dimensions
  • [general] The frames written for each event now exactly match the reported event span
  • [general] On VFR inputs, clips produced by encode mode are written at the input's average framerate: all frames are preserved, but playback speed will deviate from real time in sections whose local framerate differs from the average, and audio is not re-timed to match, so it can drift out of sync with the video (use ffmpeg or copy mode if exact playback timing is required)
  • [bugfix] quiet-mode = yes in a config file no longer breaks scans started from the app
  • [bugfix] The main window now shrinks to fit the screen on smaller displays instead of opening larger than the desktop
  • [bugfix] Fix syntax error due to f-string usage under Python 3.10 #262
  • [bugfix] Ensure motion high score is reset when thumbnails are disabled #267
  • [bugfix] Fix "stale" frame references being used as thumbnails #268
  • [bugfix] Ensure comma-separated timecode list persists in logfiles #265
  • [bugfix] Allow unknown keys in config file with a warning #242
  • [bugfix] The first frame of the video is no longer skipped by motion detection when using the region editor (-r/--region-editor) #239
  • [bugfix] Fix the Dockerfile working directory (/videos/) not matching the documented docker run command, add ffmpeg to the Docker image so the default output mode works, and update the Docker instructions #133
  • [bugfix] Input filenames containing [ or ] (or other wildcard characters) are no longer treated as patterns if the file exists. Wildcards are now also expanded in absolute paths, matches are sorted by filename so multiple inputs are concatenated in a predictable order, and nonexistent inputs are always rejected before scanning starts. #171
  • [bugfix] Fix region editor in the GUI showing the first frame of a previously removed video after changing the input list. Existing regions are kept when the new first video has the same resolution; otherwise they are reset (with a prompt to save any unsaved changes). #274
  • [bugfix] Explicitly specified -k/--kernel-size values are no longer scaled down when -df/--downscale-factor is set (auto kernel size continues to scale with the downscaled resolution) #194
  • [bugfix] Log an explicit error message if OpenCV VideoWriter fails to open an output video file #214
  • [general] Log active motion detection and event parameters (threshold, min_event_len, time_pre_event, time_post_event, merge_window, downscale_factor, frame_skip) at the start of a scan #186
  • [general] Remove the OpenCV version cap (<4.13); DVR-Scan now installs with the latest OpenCV 4.x/5.x releases #257
  • [bugfix] When the CNT subtractor is unavailable, the CLI now explains how to restore it, and the app hides the option (as well as the Fast Scan preset) instead of failing to load
  • [general] The Python package depends on opencv-python again in addition to opencv-contrib-python, as it is required by some dependencies (this reverses the v1.8.2 change from #251); both packages must be the same version
  • [general] The dvr-scan-headless package now depends on scenedetect-headless and opencv-contrib-python-headless, so it no longer installs the GUI builds of OpenCV
  • [general] New dependency on tkinterdnd2 for drag and drop support in the app
  • [general] Minimum supported Python version is now 3.11 (Python 3.10 is EOL)
  • [general] Updates to Windows distributions:
    • Python 3.12 -> Python 3.13
    • OpenCV 4.10.0.84 -> 5.0.0.93
    • PyAV 19.0.2 is now bundled #255
    • ffmpeg 8.0 -> 9.0.2

Contributors

Thanks to @elvis-epx for contributing to this release.

Check out latest releases or
releases around Breakthrough/DVR-Scan v2.0-release

Don't miss a new DVR-Scan release

NewReleases is sending notifications on new releases.

Get notifications