v3.0.0

What's New in v3.0.0

Auto-Label (Pseudo Labeling) — The Headline Feature

Tired of drawing every single bounding box by hand? v3.0.0 introduces local YOLO inference powered by ONNX Runtime. Export any Ultralytics detection model to .onnx, load it into YOLO-Label, and let the model do the heavy lifting.

• One-file setup — A single .onnx file is all you need. Class names, input size, and model configuration are all read from the ONNX metadata automatically. No separate config files, no manual setup.
• Supported models — YOLOv5, YOLOv8, YOLO11, YOLO12, YOLOv26, and end-to-end models (NMS baked in). Any current and future Ultralytics detection model that follows the same ONNX export format should just work.
• Per-image or batch processing — Click Auto Label (or press R) to detect objects on the current image, or Auto Label All to batch-process every image in the dataset.
• Adjustable confidence threshold — Use the slider to control detection sensitivity (default 25%).
• Fully editable results — Auto-labeled boxes can be moved, deleted, class-changed, or undone (Ctrl+Z) using all existing annotation tools.
• Class list mismatch warning — If the loaded model's class list doesn't match the current dataset's class file, YOLO-Label warns you and disables auto-labeling to prevent incorrect annotations.
• Skip class file entirely — If the model has embedded class names (all Ultralytics exports do), they will be loaded automatically. You can launch with just an image directory and an ONNX model.
• CLI support — Pass the .onnx file as a command-line argument: ./YoloLabel [dataset dir] [class file] [model.onnx]. Arguments are detected by file extension.
• Conditional build — ONNX Runtime is optional. Without it, the app builds and works normally — just without the auto-label feature. Pre-built releases include ONNX Runtime.

macOS Code Signing & Notarization

macOS builds are now signed with a Developer ID certificate and notarized by Apple. This means:

• No more Gatekeeper warnings — Download the DMG, open it, drag YoloLabel to Applications, and it just works. No need to right-click → Open or disable Gatekeeper.
• Inside-out code signing — All nested frameworks, dylibs, and plugins are individually signed per Apple's current requirements (macOS 13+), with hardened runtime enabled.
• Notarized on release — Release DMGs are submitted to Apple's notary service and stapled, so Gatekeeper can verify them even offline.
• Entitlements — Minimal entitlements (allow-unsigned-executable-memory, disable-library-validation) are applied only to the main executable for Qt and ONNX Runtime compatibility.

Other Improvements

• Default file dialog to application directory — File open dialogs now start at the application's directory instead of the system default, making it easier to find your datasets and models (#112)
• Bundled COCO class names — A coco_names.txt file is included in the Samples directory for quick testing with COCO-trained models (#112)

Bug Fixes

• Fixed wheel scroll on class table changing images — Scrolling the mouse wheel over the class label table no longer accidentally navigates to the next/previous image (#111)
• Fixed UTF-8 model path on Windows — ONNX model files in directories with non-ASCII characters (e.g., Korean, Chinese, Japanese) now load correctly on Windows by using std::filesystem::path for proper wide-string conversion (#113)
• Fixed dynamic-shape ONNX model handling — Models exported without fixed input dimensions are now handled correctly by reading the imgsz metadata field
• Fixed out-of-bounds read in image preprocessing — Resolved a scanline indexing bug that could cause crashes or incorrect inference on certain image sizes

Internal / CI

• Removed dead code for previously deleted save/remove buttons (#109)
• Added CI accuracy test that compares C++ inference output against Ultralytics Python reference to catch regressions
• macOS CI: inside-out code signing, Developer ID signing + notarization pipeline (#116)

Downloads

OS	File	Notes
Windows (x64)	YoloLabel-Windows-x64.zip	Unzip and run YoloLabel.exe
Linux (x64)	YoloLabel-Linux-x64.AppImage	chmod +x and run
macOS (Apple Silicon)	YoloLabel-macOS.dmg	Signed & notarized — open DMG and drag to Applications

What's Changed

• chore: remove dead code for deleted save/remove buttons by @developer0hye in #109
• fix: prevent wheel scroll on class table from changing images by @developer0hye in #111
• feat: default file dialog to application directory and add COCO names by @developer0hye in #112
• feat: add YOLO auto-labeling with ONNX Runtime by @developer0hye in #110
• fix: use std::filesystem for UTF-8 safe Windows path conversion by @developer0hye in #113
• docs: add auto-label (pseudo labeling) section to README by @developer0hye in #114
• ci: add macOS code signing and notarization by @developer0hye in #116

Full Changelog: v2.0.0...v3.0.0