microsoft/VoTTArchived
VoTT
VoTT este un software de adnotare pentru computer vision și un instrument de pregătire a seturilor de date pentru machine learning. Este o aplicație desktop concepută pentru desenarea bounding box-urilor și atribuirea de etichete obiectelor din imagini și videoclipuri, pentru a crea seturi de date de antrenament pentru modele de detecție a obiectelor.
Aplicația utilizează o interfață desktop cross-platform pentru a gestiona resursele de imagine și video. Dispune de o integrare de stocare local-first pentru a gestiona resursele media mari direct din sistemul de fișiere al mașinii gazdă și include eșantionare video controlată prin frame-rate pentru a extrage imagini specifice din fluxurile video în scopul etichetării.
Software-ul acoperă întregul ciclu de viață al datelor, inclusiv importul de resurse din stocare locală sau cloud și conversia datelor adnotate în diverse formate de machine learning prin exporturi bazate pe schemă. Include, de asemenea, criptare bazată pe token-uri pentru a securiza setările sensibile ale proiectului.
Features
- Bounding Box Interfaces - Provides a specialized graphical interface for drawing rectangular bounding boxes around objects for detection model training.
- Annotation Format Exporters - Converts internal annotation coordinates into standardized machine learning dataset formats via template-based exporters.
- 2D Object Labeling - Provides a tool for creating rectangular bounding box annotations on 2D images and video frames for object detection.
- Computer Vision Annotation - Serves as a comprehensive tool for creating high-quality annotated datasets from images and videos for vision AI.
- Data Labeling Tools - Provides interfaces and workflows for annotating datasets to support supervised machine learning tasks.
- Dataset Preparation Tools - Provides utilities for collecting, cleaning, and curating image and video samples for machine learning model training.
- Data Preparation Tools - Includes utilities to clean, format, and organize raw media assets into structures suitable for ML model ingestion.
- Dataset Exports - Exports image and video samples and their associated annotations into specific formats for AI training pipelines.
- Local-First Storage - Prioritizes direct host machine file system integration to manage large media assets without requiring server uploads.
- Image Labeling - Allows users to define object locations and assign tags within images to generate ground truth training data.
- Video Frame Annotations - Supports tagging objects across video sequences by navigating through frames to ensure consistency for ML models.
- Frame Annotation Workflows - Enables tagging of objects across video sequences by navigating through extracted frames to maintain temporal consistency.
- Media Asset Importers - Connects to local file systems and cloud storage providers to import images and videos for the labeling pipeline.
- Electron Desktop Applications - Implements a cross-platform desktop application using the Electron framework to integrate web technologies with native system access.
- Frame Sampling Strategies - Implements specific sampling strategies to extract a sequence of labelable images from video streams at set intervals.
- Canvas-Based Annotation UIs - Provides a graphical canvas interface for drawing bounding boxes and tagging objects in real-time on images and video frames.
- Desktop Shells - Uses an Electron shell to wrap the web interface, enabling native desktop window management and local file system access.
- Image Annotation - Cross-platform utility for labeling image and video assets.
- Image Annotation Tools - Tool for labeling and annotating images for object detection models.
