DLTA-AI: Revolutionizing Image Annotation
DLTA-AI is a groundbreaking tool that enhances the way users interact with image datasets. By integrating the latest in Computer Vision State-Of-The-Art models, DLTA-AI offers a smooth and user-friendly experience to simplify the process of image annotation like never before.
Installation π οΈ
Setting up DLTA-AI is straightforward. After preparing a new environment and installing PyTorch, users can easily install DLTA-AI via pip with the following command:
pip install DLTA-AI
To run DLTA-AI, simply use:
DLTA-AI
For a detailed guide on installation options and troubleshooting, refer to the Installation section in the User Guide.
Segment Anything πͺ
DLTA-AI brings annotation to new heights through "Segment Anything" (SAM), a feature that employs zero-shot segmentation for any class. SAM can enhance segmentation quality, making effective annotations even with imprecise object boundaries. Moreover, SAM is built to support both segmentation and object tracking in video mode.
Model Selection π€
The "Model Explorer" within DLTA-AI empowers users to leverage a diverse suite of models from sources such as mmdetection and YOLOv8. Users can compare, select, and download models, allowing for optimal model choice according to their needs.
Segmentation π¨
DLTA-AI provides a seamless experience for annotating single images or batches, with flexible class selection and result modification. SAM integration allows for zero-shot segmentation and quality enhancement.
Object Tracking π
Built on segmentation and detection models, DLTA-AI offers an all-inclusive object tracking solution with five models. It provides features for video navigation and customization of tracking results, including exporting to various formats and editing tracking details efficiently across frames.
Export π€
For instance segmentation, users can export their work to standard COCO format, and tracking results to MOT format or a video file. There's also an option to define custom export formats tailored to specific needs.
Other Features π
- Threshold Selection: Users can set confidence and IoU thresholds.
- Class Selection: Choose from 80 COCO classes with defaults.
- Tracking: Track specific assigned objects.
- Model Merging: Combine outcomes from multiple models.
- Customizable UI: Fully adaptable interface with light/dark theme support.
- Performance Metrics: Real-time runtime type and GPU memory usage display.
- Frame Navigation: Intuitive video navigation controls.
- Notifications: System alerts for lengthy tasks.
- User Preferences: Shortcuts and settings for personalization.
Contributing π€
DLTA-AI thrives on community contributions, welcoming involvement in bug reporting, feature requests, and code improvements. Contributors can assist through issues and pull requests.
Acknowledgements π
DLTA-AI is developed under the guidance of esteemed mentors at Ain Shams University's Faculty of Engineering, and supported by contributors who provided invaluable feedback and testing.
Resources π
DLTA-AI integrates with several resources including Labelme, SAM, and others, to offer a robust ecosystem for data annotation.
License π
DLTA-AI is available under the GPLv3 license, ensuring its accessibility and openness for community developments and enhancements.