Release notes¶
The current release is 1.9.0 (2026-10-06).
What’s in 1.9.0¶
- One installer for every machine. If the PC has an NVIDIA GPU (RTX 20-series / Turing or newer, driver 580+), training and auto-annotation run on it. If not, they run on the CPU. Nothing to configure. The training log’s first line says which you got, for example Training on GPU: NVIDIA RTX 2000 Ada Generation (16 GB) or Training on CPU (no NVIDIA GPU found). If a GPU is fitted but unused, the log says why (usually an old driver).
- Projects and folders. As many projects as you like, each with its own classes, images, review state and model versions. A project can hold more than one image set. Auto-annotation and export can be limited to one folder.
- Labelling. Bounding box (B), polygon (P) and rotated box (O). Rotate with the handle or [ ]. Pinch to zoom, two-finger scroll, Space + drag. Every shortcut is one ? key away. Copy/paste, undo/redo, autosave.
- Training. At least 50 reviewed images. Only reviewed labels train the model. Each run is a new version and warm-starts from the previous weights when classes match. Auto-annotate, review, correct, retrain.
- Export. ONNX (Standard). NVIDIA GPU / TensorRT and Hailo (Pro). Class names and pre-processing are stored in the file.
- Measurement (Pro). Length, breadth and squareness from 2D or ToF (Helios2). CSV or Excel, with a photo on every Excel row.
- App. Light theme by default (Ctrl+T for dark). Sign-in on start-up. Licence dialog shows the ctrlX WORKS steps. Training, export and licence code is compiled native — not shipped as source. Data stays in
%USERPROFILE%\AnnotationTool. - Windows only. There is no 1.9.0 macOS build.



All releases¶
| Version | Date | Highlights |
|---|---|---|
| 1.9.0 | 2026-10-06 | First commercial release. One installer (GPU or CPU); projects and folders; box, polygon and rotated-box labelling; ONNX (Standard), Hailo and NVIDIA GPU (Pro) export; measurement (Pro). |
Known limitations in 1.9.x¶
- The installer is large because the NVIDIA GPU libraries are included on every machine, whether or not it has an NVIDIA GPU.
- Training on the GPU uses full precision in this release.
- Compiling for Hailo needs the Hailo Dataflow Compiler installed separately.
- This release is for Windows. There is no 1.9.0 build for macOS.