- Corresponded to RF-DETR Core ML models. How to Train a RF-DETR Object Detection Model with Custom Data https://rectlabel.com/rfdetr_detection How to Train a RF-DETR Instance Segmentation Model with Custom Data https://rectlabel.com/rfdetr_segmentation - Exported split folders became train/valid/test. - Exported COCO file name became _annotations.coco.json. - Added "Convert pixels mask to polygon" option when exporting a COCO file.

- "Create polygon using SAM3" feature corresponds to multiple text input such as "zebra,water,tree". - When "Delete this point" for a keypoints object, RectLabel makes a backup of all annotations, deletes the point from the skeleton, and deletes the point from all keypoints objects in annotation folders. - Improved drawing speed of keypoint names, point indexes, and point coordinates, - Show label names with colored text according to the object color.

- Convert to pixels mask to polygon using the polygon smooth scrollbar There are 7 ticks on the polygon smooth scrollbar. If the pixels mask does not have the inside structure, ticks 4 - 6 are the same as ticks 1 - 3. As default, tick 3, does not consider the inside structure of the pixels mask. This is compatible with external products. 0: Convex hull 1: External contour with fewer points 2: External contour with medium points 3: External contour with enough points 4: External and inside contours with fewer points 5: External and inside contours with medium points 6: External and inside contours with enough points https://rectlabel.com/mask_to_polygon

- "Open a folder which includes a yaml file" does not add a new project when importing object names are the same as current object names to maintain color settings.

- For the Text input of the attribute type, you can input the number by such as Shift+{X}+{Y}. - Fixed the exporting problem when search.

- You can use modifier flags such as command, shift, and control for objects and attributes hotkeys. - Improved converting pixels mask to polygon algorithm.

- Corresponded to YOLO26 Core ML models. - Fixed erasing pixels using the polygon tool. - Added "Show polygon smooth scrollbar" on View menu.

- Added "Create polygon using SAM3" on Edit menu. - Added "Start SAM3 preprocessing" on Edit menu. - Added "Hide SAM3 boxes" on View menu. - Added "Filter by class" on CoreML dialog so that if you put "person", RectLabel finds only "person" class.

- Truncated, occluded, and difficult tags are saved in the YOLO format. - Create memo feature works in the YOLO format.

- When save annotations, the thumbnail on the gallery view is updated. - Escape key deselects all polygons. - Import COCO format with attributes including occluded. - Import COCO format for SA-Co/Gold and SA-Co/Silver datasets.

- Improved the Cellpose feature for gray images, for gray images, for channels, please set the same one such as (Red, Red) for better detections.

- Added "rotated box" to the Create polygon using SAM. - Improved export/import coco features so that it can export/import the track_id.

- Added "Convert to box using SAM" hotkey on Edit menu. - Shows rotated box rotation degrees at the bottom-left corner.

- Added "Create box using Tracking" on Edit menu.

- Improved Create keypoints mode so that both keypoints and the bounding box are shown and can be edited. To hide the bounding box, use View menu -> Hide keypoints boxes. - Added "Set window size" on View menu.

- Fixed the problem that the change button is not shown on the open table. - Removed the titlebar and updated the design for the macOS Tahoe.

- Improved not to crash for broken webp images. - Improved converting a pixels mask with holes to polygons.

- Added "Rename image" on File menu. - Added "Hide image name" on View menu. - Improved drawing selected multiple polygons not to show edit points.

- Added ONNX menu -> Load Cellpose ONNX Model. - Added "Export a Cellpose 16bit PNG file" option when exporting mask images. - Added "Import Cellpose 16bit PNG files" on Export menu. - Improved "Convert mask to polygon" feature so that you can use the smoothness slider to control the number of points. - Improved the speed and stability of Core ML segmentation models processed for all images. - Fixed so that when on search and changed the label name and memo, those changes are updated for the next search result. - Fixed so that when on search and relaunch app, the gallery view is correctly generated. - Fixed so that when exporting the settings file, CoreML and SAM panels are not shown.

When your YOLO text files are using the CRLF line break, RectLabel saves each YOLO text file using the CRLF line break, not LF line break.

When using "Create polygon using SAM" mode, after preprocessing the image, using the right click menu, you can use "Convert to box using SAM" feature to tighten selected existing boxes.

To save a track id for each box in the YOLO format, add an attribute named as “track id” of “Text input” type. class_index center_x center_y width height track_id 0 0.464615 0.594724 0.680000 0.769784 1 https://rectlabel.com/export/#export-yolo-txt-files

- When exporting the CreateML JSON file and training on Create ML, the error message "Data Analysis stopped without finishing" was shown if the first image has no annotations, so that RectLabel exports images with annotations first, and images witout annotations are appended to the last. - When exporting the CreateML JSON file from a single images folder, each image path included the subpath such as "person/0.jpg", but it was reverted to "image.jpg", so that you can drag & drop the CreateML JSON file into the images folder and start training on CreateML. - When converting a video file to image frames, image size was specified to 1500px, but sometimes the image was exported as 1499px, this problem was fixed. - Added "Show image full path" hotkey on View menu.

- Hidden objects are not exported. - Once opened images and annotations folders, from the second launch, you can use command line arguments to RectLabel which image files should be opened. open -a RectLabel --args -images 000000000872.jpg,000000010363.jpg open -a RectLabel\ Pro --args -images 000000000872.jpg,000000010363.jpg single folder open -a RectLabel --args -images 000000000872.jpg open -a RectLabel --args -images /Users/ryo/rcam/test_annotations/test/_test_min/images1500/000000000872.jpg multiple folders open -a RectLabel --args -images alice-pasqual-cOKmuKWSsiM-unsplash.jpg open -a RectLabel --args -images chocolate/images/alice-pasqual-cOKmuKWSsiM-unsplash.jpg open -a RectLabel --args -images /Users/ryo/rcam/test_annotations/test/_nested/donut-A/chocolate/images/alice-pasqual-cOKmuKWSsiM-unsplash.jpg nested folders open -a RectLabel --args -images alice-pasqual-cOKmuKWSsiM-unsplash.jpg open -a RectLabel --args -images donut/chocolate/alice-pasqual-cOKmuKWSsiM-unsplash.jpg open -a RectLabel --args -images /Users/ryo/rcam/test_annotations/test/_nested/nested1500/donut/chocolate/alice-pasqual-cOKmuKWSsiM-unsplash.jpg

- Added "Multi-Label Classification" option to the Export CreateML JSON file feature. - Fixed the copy/paste problem when you check on the “Auto copy from previous image” checkbox.