MANUFACIA-Vision
New Version Release Announcement

MANUFACIA-Vision version 1.1.0 has just been released.

(Installer 1.1.0.13)

Release date: July 18, 2025

Thank you for your continued use of MANUFACIA-Vision.

The changes in this version are as follows:

New features/specification changes

- Added the ability to select multiple folders in the Dataset folder dialog.

- When extracting labels from a Dataset file name, the priority for comma-separated list definitions that satisfy multiple conditions has been changed to prioritize the longest labels first, followed by the order in which the labels were registered.

- Training: The loss curve is now displayed during supervised learning.

- Training: The output layer of the feature extractor can now be arbitrarily selected during good-item learning (PATCH DISTRIBUTION).

- Added feature extractors EfficientNet B6 and B7 to the training section.

- Training: In the good sample learning (PATCH DISTRIBUTION) process, training was permitted using only good samples.

- Training: We've made it possible to select multiple feature extractors with a single algorithm and train them simultaneously.

- Training: The feature extractor can now be automatically trained on different layers using good sample learning (PATCH DISTRIBUTION).

- Inference: Heatmap rendering maintains the maximum sensitivity during training and does not update that sensitivity when performing new inferences. If the heatmap sensitivity exceeds the maximum sensitivity during model training during inference, it is ignored.

- The Model PDF report is now output as a file without displaying a preview. The file is saved in the same folder as the model, and this change removes the upper limit on the number of samples.

- Model: Added model size to the model results display and PDF report.

- Model: The ability to output good-product learning (PATCH DISTRIBUTION) and supervised learning models to ONNX format has been added. (Switchable between FP16 and FP32)

- Model: Added the ability to lower the sensitivity of the heatmap on the screen.

- The path where projects are saved can now be changed.

Bug fixes

- Fixed an issue where the width and height were reversed during the training resizing process.

- Fixed an issue where the histogram and ROC were displayed incorrectly during model inference execution.

- Fixed an issue where cached Model results were not automatically created when they did not exist.

- Inference: Fixed an issue in good-product learning (PATCH DISTRIBUTION) where the automatic batch size prediction was not underestimated.

If you would like to try our product, please contact us through our " Contact Us " page.

A representative will contact you as soon as possible.