Lite Classify 1.2

Lite Classify 1.2 can process just the area you are looking at, and adds a pixel-level pass for strips narrower than an object.

• Process This Extent Only (Tab 1): zoom to an area and work on that part alone, for a quick demo or a machine with limited memory. The whole image stays visible around it at low resolution, the source file is only read, and the extent gets its own workspace folder.
• Greenness refinement (Tab 3): after classification, re-decide two classes (such as grass and bare) pixel by pixel from Excess Green. It brings out thin green strips, and its default mode keeps patchy or dying grass whole.
• Optional DINOv3 ViT-S+/16 for Discover Classes and the classifier: use your own download from Meta. It is not included, and it stays under Meta's DINOv3 License.
• Map View (Tab 3): one Ortho / Segments / Classified selector, a segment boundaries switch that works in every view, and opacity and blend for the map layers.
• Accuracy points: Export Points to a re-importable CSV; on import, match class names, decide what happens to leftover classes, and optionally keep only points inside the processed extent; Delete Checked alongside Unassign Checked.
• Changed: class separability is now cross-validated. The earlier score could read 2.00 for nearly every pair, so values will be lower and more telling than in 1.1.
• Fixed: the minimum mapping unit sieve no longer leaves small patches at the image edge unclassified.

The manual is rebuilt on a new worked example, the 13.5 ha CAFS experimental agricultural site, with its sample data on Google Drive.

Two offline Windows installers: Setup CPU runs on any Windows 10/11 PC; Setup GPU adds NVIDIA CUDA for faster segmentation and appearance features. Installing either replaces 1.1.

User manual: https://leobarua.github.io/liteclassify/manual/

Lite Classify 1.1

Lite Classify 1.1 can make the orthomosaic itself from drone photos, and long steps can now be cancelled.

• Import > Orthomosaic from Drone Photos (WebODM): send a folder of drone photos to your own WebODM, optionally only those inside an area drawn on a map of the photo positions, follow its progress and log, and load the finished orthomosaic and DSM. WebODM is installed separately; Lite Classify includes no photogrammetry. Jobs keep running if Lite Classify is closed and are picked up on the next start.
• Cancel buttons for segmentation (the previous result is kept) and for Discover's deep features
• Fixed: Discover's segment-edges toggle works right after segmentation, before Find Groups
• Fixed: the splash screen no longer stays on top of other windows

Two offline Windows installers: Setup CPU runs on any Windows 10/11 PC; Setup GPU adds NVIDIA CUDA for faster segmentation and DINOv2 features. Installing either replaces 1.0.

User manual: https://leobarua.github.io/liteclassify/manual/

Lite Classify 1.0

First public release of Lite Classify: object-based land-cover mapping of drone
orthomosaics, offline on one Windows PC.

• Segment an orthomosaic (and optional DSM) into objects with SLIC, on CPU or NVIDIA GPU
• Discover the scene's classes with DINOv2 context embeddings and review them as image tiles
• Train a Random Forest from clicked examples and retrain until the map stabilizes
• Assess accuracy with a stratified sample: error matrix, area-weighted accuracy and
  area estimates with 95% confidence intervals

User manual: https://leobarua.github.io/liteclassify/manual/