OptimEyes Developer Adds ML Support With TensorFlow, More
OptimEyes Developer version 25.6.4.0 is now available with several major new features and improvements.
- The most exciting new development is the addition of the ML Tool, which uses TensorFlow to run Machine Learning models and perform such functions as object detection and image classification.
- The ML Tool provides a single-dialog interface for working with ML models, and integrates into OptimEyes' VisualEyes Scripting capabilities. This enables the use of ML models without complicated installation procedures or needing to write Python or C++ code.
- The Tool provides functions to help the user investigate how the ML model accepts inputs and returns outputs, helping users to navigate the often cryptic and poorly documented operations of many models.
- The Tool can store the ML model output in VES variables for use by other Tools, and can automatically overlay ROI rectangles, lines, and points on images for ML models that return object coordinates.
- On computers with CUDA-compatible nVidia GPUs, the ML Tool uses a GPU accelerated version of TensorFlow for significantly faster performance.
- The ML Tool supports the Saved Model format.
- A new Lookup Tool was created to, among other things, act as a companion to the ML Tool. An example of how they might work together is that when an ML model classifies an image, it generally returns a number representing the class rather than a text label. The Lookup Tool can be used to look up the class number in a companion CSV files and return the class label. Of course, it can also be use for other purposes where data needs to be looked up in one column of a spreadsheet based on the value in another column.
- The VisualEyes Script Tool was updated to enable variables to be set based on the values of other variables whose names are not known at the time the script was created. For example, if the ML Tool stores multiple data values in variables named "OBJECT_1", "OBJECT_2" and "OBJECT_3", VisualEyes Scripts can now iterate through them by constructing the names of these variables on the fly and looping until however many variables the ML Tool set have all been processed.
- A new Script Function was added enabling the on-the-fly generation of Calibration Objects. This can be used, for example, in conjunction with the Transform Tool to de-skew images when the amount of skew is not known at the time a VisualEyes Script is created.
- Added a way to skip the loading of individual Tools without having to renumber all of the following tools in the INI file.
- Added the option to automatically run a CustomEyes script when it is loaded rather than requiring another click to start operations.
- Documentation for several Tools was updated to provide additional guidance for how to use them when deploying CustomEyes solutions.
- The OpenCV Picture Tool was updated to automatically restore the last used backend when it is reopened.
- A bug was fixed in the OCR Tool that could cause an error when operating on an ROI that had been rotated.
- For developers of custom Tools, the order of loading include files in HL_Tool.cpp was updated to avoid errors when tool includes need to be loaded before windows.h.