Data Export Enhancements
Support for native export to COCO and YOLO formats has been added. These are standard formats widely used in computer vision, allowing users to avoid additional data conversion before use in ML pipelines.
COCO format supports two types of segmentation: RLE (run-length encoding) and Polygon (coordinates). Added the ability to include Track ID during export to maintain object connections across video frames.
These changes only affect the data structure during export and do not impact the annotation process itself.
File Name Mapping via CSV
Added the ability to change filenames during export using a CSV mapping file.
This allows control over how the dataset looks after downloading without changing the original project data.
Key benefits:
- data anonymization;
- changing file name structures;
- restoring correct frame sequences in case of incorrect naming.
Particularly useful for video where frame order is critical.
New Segment Anything 3 (SAM 3) Model
The segmentation tool has been updated with the Segment Anything 3 model for more accurate object boundary definition.
Main change: Support for text prompts. Users can describe an object in words (e.g., "person," "hands," "car"), and the system automatically creates a mask.
Currently, text prompts are supported in English only
Text Recognition (OCR)
An OCR tool has been added for automatic text extraction from images. Users can:
- recognize text in a selected area
- or across the entire image.
Recognized text is automatically added to attributes.
Translation with the help of Gemini
Translation feature: Powered by Gemini models, supporting translation between Ukrainian, English, and Russian with auto-detection.
Two modes available:
- Flash (faster)
- Pro (more accurate for complex contexts).
Additional translation rules can be set via the Memory (context) field.
Timeline for Video Annotation
A Timeline panel has been added for video processing.
This panel allows you to see in which frames each object appears, as well as how its attributes change over time. This significantly simplifies navigation and video analysis.
Through the Timeline, you can modify attributes for a whole range of frames at once, without needing to open each frame individually.
Object filtering is also available, which makes working with large scenes easier.
Skeleton Annotation Improvements
The ability to display point (vertex) names for skeleton and mesh object types has been added.
The user can choose a light or dark label display
mode depending on the image background. This
simplifies annotation review and reduces the number
of errors when working with complex structures.
Multi-line Text Attributes
The ability to enter multi-line text in text attributes has been added. To enable this, simply check the checkbox on the project configuration page next to the corresponding “Editor behavior” option.
This improves the readability of long values and simplifies working with large amounts of text.
Report Updates (File Stats)
In the Project File Statistics section, a new column “Current Stage” has been added to the table.
It displays the current status of a file (e.g., Annotation, Verification, Completed) and allows filtering data by stages.
This significantly simplifies project progress analysis and productivity calculation without the need for additional data processing.
Made with Bullet