X-AnyLabeling v3.2.6

DEIMv2.mp4

Baidu Cloud: https://pan.baidu.com/s/1pgaw02inCvbEgOme9ajDJA?pwd=e528

🚀 New Features

• Add support for using backspace key to delete the last vertex when creating polygon and line shapes (#1151)
• Introduce DEIMv2 real-time object detector
• Add the ability to process all images at once with the Florence-2 model (#1152)
• Add max_det parameter for maximum detections in YOLO model (#1142)

🐛 Bug Fixes

• Enable auto_use_last_gid for digit shortcuts and reset on image switch (#1149)

🛠️ Improvements

• Add troubleshooting steps for exporting empty Mask images (#1153)
• Update rotation increments to radians and format degree display

🌟 Contributors

A total of 4 developers contributed to this release.

Thank @lhj5426, @sckiyo, @Vlad188-1, @CVHub520

Full Changelog: v3.2.5...v3.2.6

X-AnyLabeling v3.2.5

Baidu Cloud: https://pan.baidu.com/s/1pgaw02inCvbEgOme9ajDJA?pwd=e528

🚀 New Features

• Add Group ID Manager feature with Ctrl+Shift+G shortcut for auto-using last group ID (#1143)

🐛 Bug Fixes

• Fix image folder loading delays by implementing async EXIF detection (#e4bf7f6)

🛠️ Improvements

• Update Group ID Manager section in user guide (#1146)
• Add --qt-platform argument for improved performance on Fedora KDE environments (#1145)

🌟 Contributors

A total of 2 developers contributed to this release.

Thank @vodnikss, @CVHub520

Full Changelog: v3.2.4...v3.2.5

X-AnyLabeling v3.2.4

image_classifier_demo.mp4

Baidu Cloud: https://pan.baidu.com/s/1pgaw02inCvbEgOme9ajDJA?pwd=e528

🚀 New Features

• Introduce a dedicated multi-class image classifier with a streamlined workflow (#480)
• Add support for Ultralytics image classification tasks
• Add support for deleting group IDs from objects (#1141)
• Add checkboxes for description and label visibility control (#1139)
• Add loop select labels functionality for sequential shape selection (#1138)
• Add support for drawing rectangle shapes outside canvas with auto-clipping (#1137)
• Add radio button support for attribute selection (#1135)
• Add an option to skip empty label files when creating YOLO datasets and update the UI (#1131)
• Add an option to preserve existing annotations when uploading YOLO labels (#1125)
• Add custom provider support for the chatbot and enhance the model dropdown feature
• Add cross-widget reference support in VQA dialog using @widget_title syntax
• Add support for paths wrapped in quotes for models and data
• Add select/deselect all shapes feature (#1092)
• Implement keyboard shortcuts for image navigation (A/D)

🐛 Bug Fixes

• Resolve inconsistent attribute behavior after shape creation and switching (#1134)
• Ensure linestrip's vertex is drawn regardless of selection state (#1134)
• Resolve issue where CUDA device count returns 0 after model export in Ultralytics training (#1126)
• Fix Windows path separator error in train_script.py
• Fix inconsistent shape order when using existing shapes for recognition in PP-OCR
• Fix issue where group ID info was only updated after successful modification

🛠️ Improvements

• Auto-update the attributes panel after shape creation (#1134)
• Improve shape selection logic for point and line types to enhance user interaction (#1134)
• Enhance shape selection for partial re-recognition in PP-OCR (#1113)
• Enhance AI prompts in VQA with cross-component and annotation data references
• Preserve original shape properties when skipping detection in OCR (#1116)

🌟 Contributors

A total of 3 developers contributed to this release.

Thank @sckiyo, @Vlad188-1, @CVHub520

Full Changelog: v3.2.3...v3.2.4

X-AnyLabeling v3.2.3

Baidu Cloud: https://pan.baidu.com/s/1pgaw02inCvbEgOme9ajDJA?pwd=e528

🚀 New Features

• Implement EXIF orientation processing for images during import, with user feedback and backup features
• Add mask fineness control slider for SAM series models to adjust segmentation precision (#1114)
• Add Re-recognition feature for PP-OCR models (#1113)
• Add support for PP-OCRv5 model (#1113)
• Add copy coordinates to clipboard feature
• Add circle to polygon transform feature (#1112)
• Add Navigator feature for high-resolution image navigation and zoom control (#1102)
• Add refresh button to sync data with main window and display success popup for VQA Dialog
• Enable shapes field export for VQA Dialog

🐛 Bug Fixes

• Fix 'gbk' codec decode error on windows during Ultralytics traning (#1115)
• Prevent deleting label file when "Keep Previous Annotation" is enabled by showing a warning to disable it first
• Fix Chinese character path support in crop image saving
• Enable ONNX Runtime library linking in x-anylabeling-*-gpu.spec
• Fix first image showing old dataset data after refresh in VQA dialog

🛠️ Improvements

• Enhance window title to display current image progress (#936)
• Adjust label bounding box calculations for improved text positioning
• Improve circle label positioning to center display
• Streamline color retrieval by using parent method for RGB values in LabelModifyDialog
• Update field count in VQA ExportLabelsDialog for accurate table height calculation and make table items non-editable
• Enhance button styles and dialog layout for improved user experience
• Enhance Ultralytics train/val dataset split with true randomization (#1095)
• Optimize the experience of adjusting the shape (#1094)

🌟 Contributors

A total of 6 developers contributed to this release.

Thank @Einstellung, @lhj5426, @sckiyo, @xiaomaxiao, @zhixuwei, @CVHub520

Full Changelog: v3.2.2...v3.2.3

X-AnyLabeling v3.2.2

VQA-v2.mp4

Baidu Cloud: https://pan.baidu.com/s/1pgaw02inCvbEgOme9ajDJA?pwd=e528

🚀 New Features

• Support batch editing for multiple shapes (#1084)
• Introduce AI Assistant for VQA
• Add prompt template management to VQA
• Enhance VQA dialog with a new UI including sidebar toggles, streamlined navigation controls, improved page navigation, loading indicators, and updated button styles

🐛 Bug Fixes

• Fix issue with dragging and moving the image (#1088)

🛠️ Improvements

• Optimize SAM inference memory management (#1086)

🌟 Contributors

A total of 4 developers contributed to this release.

Thank @jdobbang, @zhaoruibing, @zhixuwei, @CVHub520

Full Changelog: v3.2.1...v3.2.2

X-AnyLabeling v3.2.1

Baidu Cloud: https://pan.baidu.com/s/1pgaw02inCvbEgOme9ajDJA?pwd=e528

🚀 New Features

• Add support for showing/hiding shape attributes on the canvas (#1076)
• Add functionality to save training logs with timestamp upon dialog closure (#1077)

🐛 Bug Fixes

• Skip validation for auto-labeling special constants
• Prevent closing UltralyticsDialog during active training session (#1077)
• Improve WSL2 detection for image file handling in UltralyticsDialog (#1077)
• Add UTF-8 encoding to file opening in validate_data_file function (#1077)
• Resolve Windows multiprocessing and matplotlib segfault issues

🌟 Contributors

A total of 2 developer contributed to this release.

Thank @FreemanTang, @CVHub520

Full Changelog: v3.2.0...v3.2.1

X-AnyLabeling v3.2.0

UltralyticsTrainingPlatforms.mp4

Baidu Cloud: https://pan.baidu.com/s/1pgaw02inCvbEgOme9ajDJA?pwd=e528

🚀 New Features

• Introduce Auto Training Platform for Ultralytics tasks (Detect/Segment/OBB/Pose)

🌟 Contributors

A total of 1 developers contributed to this release.

Full Changelog: v3.1.2...v3.2.0

X-AnyLabeling v3.1.2

AMD.mp4

Baidu Cloud: https://pan.baidu.com/s/1htRqSLCsJOgbSs-0R4OMBA?pwd=j3yz

🚀 New Features

• Introduce Auto Mask Decode (AMD) mode for continuous point tracking (#1060)
• Add new RF-DETR models (medium, small, nano) and fix input width typo in configuration files (#1069)
• Enhance range selection for group ID modification with new input fields and validation (#1035)
• Add support for MM-Grounding-DINO annotations upload

🐛 Bug Fixes

• Update error message for label validation to specify 'exact' in config file (#1064)
• Fix issues when drawing rectangular boxes (#1063 by zhixuwei)
• Add try-except for mask_to_polygons for supervision version compatibility (#1055 by adarshs)
• Improve segmentation handling by filtering invalid entries and avoiding duplicate points in polygon mode (#1032)
• Resolve KeyError when importing files via drag and drop (#1030)
• Enhance image saving logic to handle non-ASCII paths and improve multiprocessing handling in frozen environments (#1021)
• Improve crop region validation and handle empty cropped images with warnings (#1021)
• Emit model_loaded signal even if loading custom model configuration fails
• Update frame ID extraction logic to handle underscores and non-digit cases (#1020)

🛠️ Improvements

• Correct variable name from 'has_vasiable' to 'has_visible' for accurate keypoint processing
• Simplify toggle button text for clarity in label and shape information display
• When looping through shapes, display their fill colors (#1025 by zhixuwei)

🌟 Contributors

A total of 4 developers contributed to this release.

Thank @zhixuwei, @adarshs, @CVHub520

Full Changelog: v3.1.1...v3.1.2

X-AnyLabeling v3.1.1

video.mp4

Baidu Cloud: https://pan.baidu.com/s/16ESoJxcIcmw2DKKs06sMuA?pwd=vsnx

🚀 New Features

• Add customizable field export options for VQA dialog
• Add ability to adjust the visible area of the image by dragging the mouse (#1019)

🐛 Bug Fixes

• Fix VQA keyboard shortcut (Ctrl+Q → Ctrl+1)

🌟 Contributors

A total of 2 developer contributed to this release.

Thank @zhixuwei, @CVHub520

X-AnyLabeling v3.1.0

VQA.mp4

Baidu Cloud: https://pan.baidu.com/s/1K5PC-BUclSnJjEDNxQVY1g?pwd=f4am

🚀 New Features

• Support RMBG v2.0 model for image matting
• Add output_path parameter to COCO label converter methods for custom output paths
• Add real-time result preview for matting and depth estimation tasks
• Add GUI support for uploading custom label classes (#988)
• Add rectangle scaling and edge adjustment with mouse wheel support (#989)
• Add automatic update check on startup
• Add Visual Question Answering tool

🐛 Bug Fixes

• Improve error handling and logging for annotation export and upload processes (#974)
• Fix annotation_id increment in COCO data processing (#976)
• Fix failure to click again after custom model loading
• Fix scrollbar slider display issue
• Fix issue where copied shapes fail to be saved
• Fix auto-save bug after undo operations when switching images (#1013)

🛠️ Improvements

• Add solution to CUDA dependency error: Could not locate cublasLt64_12.dll. Please make sure it is in your library path! (#1014)
• Add solution to efficiency improvement plan for multi-object keypoint annotation and grouping (#982)
• Add CLA, contributing templates, and README contributor section
• Improve QWebEngineView import error handling in chatbot
• Improve thumbnail rendering by mapping file extensions to model types in auto-labeling service
• Improve shape adjustment convenience
• Improve click-to-move editing with state cleanup and cursor feedback

🌟 Contributors

A total of 7 developer contributed to this release.

Thank @1955946542, @donkinone, @ljh725, @pipihuang2, @sunmooncode, @zhixuwei, @CVHub520

Full Changelog: v3.0.3...v3.1.0