Release Notes
Version 2.25.3.0 Update Notes
Updated the rotated object detection model
Updated the hybrid model
Updated the Sky Eye system and the video stream detection system
Fixed several issues that could cause training failures
Version 2.25.2.0 Update Notes
Updated ultra-high-resolution training mode for supervised defect segmentation; now supports training with images resized up to 3584×3584 resolution
Updated pixel-level training for unsupervised defect segmentation: now only one normal sample is needed to train and detect unknown defects effectively; removed the 300-image dataset limit
Removed the full-image mode in unsupervised segmentation; now all detection is region-based. For full-image detection, draw the detection region as the entire image
Version 2.25.1.0 Update Notes
Updated SDK encryption method: all SDKs now require a hardware dongle for use
Fixed issues causing training failures in some models
Added support for local HTTP inference in C#