MusaMahmood/mcap
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases 1 Wiki Activity

1 Release 1 Tag

RSS Feed
  • v0.1.0 9ff30ea1d3
    Compare

    mcap 0.1.0: videos, and what's in your images Stable

    MusaMahmood released this 2026-10-10 03:46:36 +00:00 | 43 commits to master since this release

    The first release of mcap: a library of screenshots, photos and videos, each kept as one .mcap file, read with OCR, labelled, segmented and searchable.

    What's new

    Videos and animated GIFs

    • Import: MP4, MOV, MKV, WebM, AVI, WMV, MPEG and FLV import like photos (via FFmpeg). Each is stored whole, with a poster frame for the grid, OCR and labels. The date, device and location come from the video's own metadata.
    • Grid and Capture window: thumbnails show a play mark and duration (videos) or GIF. Play in the default player opens the original.
    • Any size: videos are copied a few MB at a time, never read whole into memory. This needs .mcap file version 4 (versions 1–3 still open).

    Segmentation: what's in each image

    • Choose folders: the Segments window has a checkbox per library folder; nothing is segmented unless you tick its folder. Uses Mask2Former (Swin-B, 133 COCO classes): people, animals, things, and sky, wall, grass and so on.
    • Search: has:dog, has:"dining table", has:person has:"cell phone".
    • Stats: What is in the images lists how many images hold each class; click one to search for it.
    • Capture window: segments drawn over the image. Opacity, a legend to hide classes, colours you can pick, and hover for class and score.
    • Speed: about 17 images a second on an RTX 5080 laptop, about 5 KB stored per image. -segment and -erase segments on the command line.

    Also

    • Shutdown: closing the app says what it's waiting on, and stops an OCR scan mid-batch instead of finishing it.
    • Fixes: a crash importing videos of some frame sizes, and a crash on exit.

    Download

    mcap-0.1.0-windows-x64.zip, for Windows 10 and 11 (x64). Unzip anywhere and run mcap.exe; nothing to install.

    • Size: 2.2 GB zipped, 3.0 GB unpacked.
    • Included: onnxruntime, FFmpeg, the OCR, label and segmentation models, and NVIDIA's CUDA 13 runtime and cuDNN 9.
    • GPU: on an NVIDIA GPU (RTX 20 series or newer) with a current driver, OCR, labels and segmentation run on the GPU. Otherwise they run on the CPU, which is slower but works the same.
    • Library folder: the library defaults to D:\McapIndex\mcap. Point it elsewhere with -library "E:\My library" (see README.txt).

    SHA-256: 85bd663b34cc5e6a6ef560b926744de3ff3195df5637fc9c48c7aecdec7c9f10

    Downloads