Keypoints Studio v0.3.0. Action Recognition Improvements

Overview

This release introduces significant improvements to the action recognition and object detection annotation workflow, focusing on precision editing, usability, and YOLO-compatible dataset generation.

---

New Features

Bounding Box Editing System

• Implemented 4-corner interactive resizing for precise geometry control
• Added double-click and context-menu based box editing

Annotation Workflow Improvements

• Improved sidebar architecture for navigation and autosave support
• Enhanced dataset browsing across structured image directories

Dataset Export Pipeline

• Full YOLO detection format export (.txt)
• Automatic generation of classes.txt for training reproducibility

Clipboard System Enhancements

• Single and batch geometry paste support
• Enables fast annotation replication across frames

Keyboard Interaction Model

• W — Toggle draw mode
• A / D — Image navigation
• Ctrl + S — Save annotations
• Delete — Remove objects
• Ctrl + F — Fit view

---

Technical Notes

• Optimized for Ultralytics YOLO training pipeline compatibility
• Improved annotation throughput for large-scale dataset creation
• Reduced interaction overhead in manual labeling workflows

---

Impact

This update improves annotation efficiency and transforms Keypoints Studio into a more robust dataset creation tool for computer vision and machine learning pipelines.

Keypoints Studio v0.2.0 — Pose + Action Recognition labeling

Overview

Keypoints Studio is a desktop tool for building YOLO pose datasets with a PySide6 GUI (Ultralytics YOLO), plus an optional Action Recognition mode for bbox + class labeling.

Highlights

Pose / keypoints

• Auto-annotate folders and preview/edit bboxes & keypoints
• Custom keypoint schema (reorder/filter YOLO keypoints + add custom slots)
• Save labels next to images (image.jpg → image.txt)
• Labeled-folder mode: if keypoints.txt exists, GUI loads existing labels for preview (no model inference) and disables auto-annotate for that folder
• Root directory workflow: browse multiple child folders; Next/Previous folder navigation
• Annotation_Done: when advancing folders, finished folders (with keypoints.txt) can be moved under Annotation_Done/
• Productivity: Auto save, Ctrl+V (bbox only from previous frame), Ctrl+B (keypoints only), W crosshair bbox drawing, global A/D navigation (Pose mode)

Action Recognition (labelImg-style)

• Switch AR screen for rectangle annotations + class selection
• Classes via data/ar_classes.txt
• Saves YOLO detection .txt per image + writes classes.txt

Configuration

• data/custom_classes.txt for extra custom keypoint slots (loaded at startup)

Requirements

• Python + Conda recommended
• See README.md for install/run instructions

Notes

• Do not commit large .pt model weights to git (use .gitignore).

Full documentation

See the repository README.md.

Keypoints Studio v0.1.0

Keypoints Studio v0.1.0 introduces a PySide6 GUI for YOLO pose dataset annotation and editing, with custom keypoint mapping, autosave navigation, and labeled-folder viewing mode. Outputs YOLO-pose .txt labels next to images