v6.0.0 Release Notes

New Features

YOLO Object Detection Models

MIPAR now supports YOLO object detection models alongside existing SegNet and U-Net segmentation models. While segmentation models classify every pixel in an image, YOLO models identify and locate discrete objects with bounding boxes – ideal for counting particles, detecting defects, or any application where you need to find and analyze objects. Training a YOLO model uses the same familiar workflow: trace your features in the AI Session Processor, select YOLO as the Model Type, and click Train. Your traced masks are automatically converted to bounding box training data. A key difference – YOLO models can be trained with just a single class, unlike segmentation models which require at least two. When tracing for YOLO, be sure to mark all visible instances of each object class in your training images. The Apply Model interface adapts when a YOLO model is loaded, displaying detected objects as labeled bounding boxes overlaid on your image. A new Confidence Threshold slider lets you filter detections in real-time – lower the threshold to catch more objects (including uncertain detections), or raise it to show only high-confidence results. An Object Count display updates as you adjust the threshold, giving immediate feedback on detection performance. YOLO is best suited for counting discrete objects rather than measuring precise boundaries. For applications requiring both detection and detailed measurements, YOLO outputs can be fed to Spotlight to create a two-stage detection → segmentation pipeline. Read the comprehensive Choosing a Model Type guide to learn more about selecting the right architecture. See Advanced Settings and Apply Model for detailed parameter documentation.

Function Search in Image Processor

The Image Processor now includes an integrated search system that lets you locate recipe functions by typing keywords instead of navigating through menus. Start typing to filter available functions in real-time, then select your desired operation directly from the search results. This is particularly useful when you know what you want to do but can’t remember which menu contains the function, as a power user who finds keyboard interactions more efficient than mouse clicks, or when exploring MIPAR’s extensive processing library for operations you haven’t used before.

NVIDIA Blackwell GPU Support

MIPAR v6.0 delivers full support for NVIDIA’s Blackwell architecture, including the RTX 5000-series GPUs. Deep Learning model training and application, as well as Spotlight segmentation, now take full advantage of the latest GPU hardware. If you’re running Blackwell-based hardware, ensure your NVIDIA drivers are up to date and MIPAR will automatically utilize your GPU for accelerated processing. See Deep Learning System Requirements for more information on supported hardware. First Launch: GPU Compatibility Setup The first time you launch MIPAR v6.0 with a newer GPU (including RTX 5000-series), you will be prompted to enable GPU compatibility. If you choose “Yes,” MIPAR will perform a one-time compilation process. This may take up to 15 minutes the very first time per system user, but subsequent launches will only take a few seconds. This compilation is required to enable GPU acceleration for Deep Learning and Spotlight on newer hardware. Known Issue: U-Net Training Performance Training U-Net models on very modern GPUs (those requiring forward compatibility) currently takes significantly longer than expected. If you experience slow U-Net training times, consider using a SegNet model instead, which does not have this limitation. Alternatively, if you have multiple GPUs available, you may select an older GPU for U-Net training. This performance issue is expected to be improved in future beta versions.

Major Performance Improvements

We’ve significantly enhanced MIPAR’s performance across the application. Deep Learning model application, measurement generation, and UI responsiveness have all been accelerated – in many cases by orders of magnitude. Key improvements include: Accelerated Deep Learning model application tiling for faster inference on large images with reduced tile boundary and edge artifacts. Accelerated the mapping of custom measurements to other values. Accelerated printing measurements to file. Accelerated the opening of windows, especially those that don’t display images. Accelerated recipe execution in the Image Processor. Accelerated path length measurements, especially when there are a large number of features in an image. Accelerated the generation of label and color-by-measure overlays. Accelerated the adding of buffer to images during processing. Accelerated the identification of features in images during processing. Accelerated response when opening files with drag-and-drop.

Improvements

Base Package

• Added “Nearest Edge Distance” under Math > Replace With, which draws lines between feature edges (perimeter-to-perimeter) rather than centroids. “Nearest Distance” renamed to “Nearest Centroid Distance” for clarity.
• Improved text-mapped measurement reports in Color-by-Measure by keeping the Configure Report button always visible and correctly handling overlay and histogram options for text values.
• Smart Erode/Dilate improved edge feature filtering by introducing alternative padding options. Image can be padded with zeros, ones, replicate or mean image pixel value.
• Toggling layer visibility no longer deselects the active layer row.
• Implemented a file verification step at the end of a batch process, to confirm files are ready to read if saved to network storage.
• Improved user warning when Report Generator feature is available, but Microsoft Word is not installed.
• Improved complex custom measurement mapping when the output values are heterogeneous and contain text
• Improved handling of the reference image used for recipes steps when crop or resize steps are present.
  

Deep Learning Extension

• Proposed YOLO tile grid in Session Processor now auto-computes from image size and model variant to match the network’s ideal input resolution, replacing the previous hardcoded 2×2 default.
  

Spotlight Extension

• Added P/B keyboard shortcuts to switch between Point and Box modes when the Snap tool is active.
  

3D Extension

• 3D Toolbox, added instructions text to the Manual Edit tool for using 2D Wand, 2D Selection and 3D Selection tools.
• Improved the resolution of the reconstruction images when exported out of Animate 3D Visualization app.
  

Bug Fixes

Base Package

• Fixed bug with Manage Layers that would prevent layers from being moved if the recipe had Image Measurements assigned to any of the Layers.
• Fixed bug in Load Companion image that caused logical images saved in 8-bit to load as grayscale on recipe load but as logical as a new recipe step.
• Fixed bug in the Session Processor Intercepts>Rotating measurement that would allow the parameters windows to be accepted without filling out all of the required settings.
• Fixed bug with in Calibrate Scale window that prevented ‘Auto’ from working on binary images.
• Added a buffer for the Report Generator file store to help reduce issues with slow storage or network disruptions when generating files from the Batch Processor.
• Fixed bug with waitbar handling that would prevent Dark and Bright Texture filtering from working.
• Fixed bug with some feature output printing during a batch process.
• Fixed bug generating report from a logical custom measurement in the Batch Processor.
• Fixed error printing report when a mapped measurement is a mix of text and numbers.
• Fixed bug with running recipes with custom measurement mapping on text measurements and prevented this condition going forward.
• Colormap selector will now preview the currently selected colormap and text in the dropdown.
• Fixed bug where pasting in edit fields in Image Processor incorrectly triggering recipe paste when the recipe clipboard had content.
• Improved how MIPAR handles extremely long file paths on Windows.
• Fixed some instances of a toolbar showing up in the image overlay.
• Fixed bug where crop doesn’t properly set when switching Layers view.
• Improved handling of the reference image used for recipes steps when crop or resize steps are present.
• Fixed wait bar not updating during measurement generation in the Session Processor.
• Fixed bug that would reset custom Report Generator templates on launch.
• Submit Feedback Window: Improved handling of large log files by setting a maximum character count and removing filler characters.
• Fix bug preventing the printing of reports from resumed sessions.
• Prevent silent crash when MIPAR encounters an inaccessible Python installation on the system PATH environment variable.
• Fixed bug in Image Processor where zoom would reset on every recipe jump if at least 1 Resize Image or Crop Image step was above target step
  

Deep Learning Extension

• Fixed YOLO tiled training producing poor inference results by matching inference tile grid to training scale, and fixed training failures when tiles contained no objects.
• Fixed an error when dragging and dropping a Deep Learning model into the Image Processor without an image loaded.
• Fixed error when editing some Call Deep Learn recipe steps.
• Proposed YOLO tile grid in Session Processor now auto-computes from image size and model variant to match the network’s ideal input resolution, replacing the previous hardcoded 2×2 default.
• Fixed YOLO tiled training producing poor inference results by matching inference tile grid to training scale, and fixed training failures when tiles contained no objects.
• Fixed bug with training YOLO models due to message outputs not properly initializing.
• Fixed bug with applying YOLO models due to missing dependencies.
  

Spotlight Extension

• Fixed bug in Spotlight that would prevent setting Custom prompts when the Companion image does not have nay features.
• Fixed occasional error when opening the Session Processor after upgrading MIPAR.
• Fixed error when switching between using Spotlight or the Snap tool, training a deep learning model, and back.
• Fixed slowdown in DL training that occurs after annotating with the Snap tool.
  

3D Extension

• 3D Toolbox fixed bug with closing the Animate 3D Visualization window that would print a console error.