Zastra vs. RectLabel

We are set to enhance the platform by adding annotation capabilities specifically designed for segmentation tasks. The Zastra repository will feature cutting-edge algorithms that streamline segmentation processes, thereby improving active learning across various datasets. To support this, we will implement complete end-to-end machine learning operations, which will include dataset and experiment version control, as well as templated pipelines that facilitate model deployment in standard cloud environments and on edge devices. By embedding advancements in Bayesian deep learning within the active learning framework, we aspire to significantly enhance overall performance. Additionally, we will improve the precision of annotations by utilizing specialized architectures like Bayesian CNNs, which will ensure superior outcomes. Our dedicated team has committed considerable time and resources to create this innovative solution tailored to your requirements. While we are consistently adding new features and improvements to the platform, we invite you to participate in a trial run soon! Zastra offers a variety of essential functionalities, such as active learning for object classification, detection, localization, and segmentation across multiple formats, including images, videos, audio, text, and point cloud data. This versatility not only makes Zastra a powerful tool for addressing a wide range of data challenges, but it also positions us well to adapt to future needs in data processing and machine learning advancements.

An offline image annotation tool is designed to support both object detection and segmentation tasks effectively. Users can create various shapes such as polygons, cubic bezier curves, line segments, and points for accurate labeling of images. It also enables the drawing of oriented bounding boxes, particularly useful for aerial imagery analysis. Additionally, the tool allows users to label key points that can be interconnected by skeletons and offers the capability to paint pixels using brushes or superpixels. It ensures compatibility with different machine learning formats by supporting both PASCAL VOC XML and YOLO text reading and writing. Furthermore, users have the option to export their annotated data to CreateML for object detection and image classification tasks, as well as to COCO, Labelme, YOLO, DOTA, and CSV formats. The tool accommodates diverse project requirements by enabling the export of indexed color mask images and grayscale mask images. Users can easily modify settings related to objects, attributes, hotkeys, and fast labeling features to enhance their workflow efficiency. A customizable label dialog allows for smooth integration with attributes, and one-click buttons streamline the selection of object names. With an impressive auto-suggest feature that considers over 5000 object names, users can search for objects, attributes, and image names conveniently in a gallery view. Automatic labeling is facilitated through Core ML models, and the tool includes OCR technology for automatic text recognition. It also offers features to convert videos into image frames and execute image augmentation tasks. Language support covers English, Chinese, Korean, and 11 additional languages, thus catering to a wide-ranging user base and boosting productivity across various regions.