F-Annote

Annotate · Train · Deploy

Changelog

Release history

Every version, every change. All v2.x updates are free for existing buyers.

v2.0.0LatestJuly 2026

Full rewrite from scratch. The jump from V1 to V2 is not an update - it's a different product.

Annotation canvas

  • Keyboard-first workflow - draw, label, navigate without ever leaving the keyboard
  • Smooth pan & zoom with mouse wheel and middle-click drag
  • 30-step undo / redo
  • Unannotated image filter to find unlabelled images instantly
  • Class-colour coding with auto-assigned palette

Training pipeline

  • YOLOv8 & YOLO11 support (nano → xlarge)
  • Configurable epochs, batch size, image size, patience
  • Real-time training progress with live loss graph
  • Auto train/val/test split
  • One-click model export to TensorRT .engine

Auto-annotate

  • Single-image auto-annotate with confidence threshold slider
  • Batch auto-annotate across an entire project folder
  • Works with any custom-trained YOLO model
  • COCO pretrained base model mode - annotate without training first (yolov8n/s/m, yolo11n/s)
  • Per-class COCO mapping - map your project classes to COCO categories, skip the rest
  • Model cache - repeat auto-annotate is near-instant, no reload between presses

Inference & screen capture

  • Live screen capture with configurable capture region
  • Real-time inference overlay on screen
  • Supports both .pt (PyTorch) and .engine (TensorRT) models
  • Global hotkey (P) to toggle capture - works while another app is focused
  • COCO base models available in Test Detection (yolov8n/s/m, yolo11n/s - auto-downloaded)

Export formats

  • YOLO (ZIP with images + labels)
  • COCO JSON
  • Pascal VOC XML
  • CSV (flat label list)
  • Configurable train/val/test ratios per export

Reliability & distribution

  • Projects stored next to F-Annote.exe - survive app updates without losing data
  • Crash logger - unhandled exceptions written to crash.log for easy bug reports
  • Self-contained bundle (~4.5 GB download, ~6.7 GB unzipped) - no Python, no pip, no setup required
  • Gumroad license activation on first launch (frozen exe only)
v1.0Legacy2025

The original proof-of-concept. Basic bounding-box annotation only, no training, no export variety.

Features

  • Basic bounding-box drawing on images
  • Simple class list
  • YOLO-format label export only
  • No training, no auto-annotate, no screen capture

Have a feature request? Let me know →