Large image and TIFF support for satellite and aerial imagery

Satellite and aerial pipelines run on TIFFs that are too big for a general-purpose image viewer. Multispectral, float32, often hundreds of megabytes, sometimes more pixels in a single tile than a desktop monitor will ever display. v5 either refused these files outright or wedged itself trying to decode them in one pass.

LabelMe v6 loads them, displays them, and lets you annotate at any zoom level.

LabelMe v6 annotating a large aerial TIFF image at 3% zoom with field, building, and bus annotations

26,456 x 22,892 px, 136 MB. Aerial image from OpenAerialMap.

What changed

Decoding moved out of the UI thread. Opening a large file no longer freezes the window. The image streams in while the app stays responsive, so you can start panning before the full decode finishes.

Float32 and multispectral TIFFs are first-class. GeoTIFF reflectance bands, elevation rasters, and other non-uint8 data load directly. No prior conversion to PNG, no precision loss from clamping into 0–255.

Pan and zoom scale to the file. Drawing a polygon vertex at 3% zoom on a 26,000-pixel-wide canvas behaves the same as drawing on a 1080p photograph. The coordinate space the annotation lives in is the full image, not the on-screen viewport.

When to reach for it

Satellite tiles. Sentinel, Landsat, commercial imagery: anything you'd otherwise tile down before annotating.
Drone orthomosaics. Stitched aerial captures that are big in pixels but routine to inspect at full resolution.
Slide scans and panoramic captures. Anything where the interesting object is a small fraction of a very large frame.
Multispectral or elevation rasters. When the underlying data isn't 8-bit RGB and you'd rather not pre-convert.

If you only ever annotate phone-camera photographs, this isn't a feature you'll notice. For everyone working with imagery that doesn't fit on a screen, it's the difference between using LabelMe and not.

Where to find it

Update to LabelMe v6 or later from the download page. No setting to flip; large-image support is on for every file you open.

See the v6.0 release notes on GitHub for the full changelog. The decode-on-a-worker-thread change landed in #1811; multispectral and float32 TIFF support went in via #1812.

If a particular file refuses to open, file an issue with the format and a sample if you can or ask on Discord.