Industrial AI vision software platform AIDI V2.3 is officially released!
AIDI team actively listened to users' feedback and spent nearly a year on research and development, adding and optimizing 30+ features based on the AIDI 2.2 version, and finally delivered the much-awaited AIDI 2.3, which successfully made its debut at the just concluded SIAF 2022 in Guangzhou. Being more powerful, easier to and with more intuitive data, AIDI2.3 has attracted many visitors and won the praise of many industry customers at the show.
About AIDI
AIDI is a deep learning-based industrial vision software platform that empowers a wide range of industrial applications, effectively solving problems such as location recognition, classification and character recognition of complex defects, with strong compatibility. AIDI has a wide range of built-in application modules that require no programming and help users quickly build and iterate models to meet the differentiated needs of different business scenarios helping the industry upgrade intelligently.
The software mainly includes four core functional modules (Location, segmentation, inspection and classification) with an OCR character recognition toolkit.
In this AIDI2.3 version of the all-round full upgrade, these major highlights should not be ignored:
1. Unsupervised segmentation - fully upgraded modules
For scenarios where NG samples are not easy to obtain but OK samples are easy to obtain, in which a certain number of OK samples are required, AIDI will identify the differences between the image and OK samples by learning to detect the defects in the defective map, providing a solution to the problem of small samples for industrial defect detection.
Applicable scenarios:
Uniform conditions such as target form, image features, and environment.
Periodic texture.
2.Upgraded Pinpointing - template function
Improve the speed of the original node template matching.
Applicable scenarios:
The approximate shape of each detected target object can be displayed.
Adding a new layout module for assembly inspection scenarios.
Applicable scenarios:
Check whether the specified area has a specified number of inspection targets.
3. Upgraded OCR-module function
Improve the speed of original node template matching, and it is able to match any shape string.
Applicable scenarios:
Strings are not arranged in a straight line.
Adding new string template, which will organize characters in a specified direction as a string output; support regular expression lookup.
Applicable scenarios:
Longer strings / unequal widths / arranged in a straight line
4. Segmentation-inference results support post-processing
1) Support the results of the segmentation inspection into rectangular boxes.
Applicable scenarios:
When the length and width of the inspection target are fixed, the test result can be converted into a fixed-size rectangular box to facilitate the processing of subsequent modules and ensure the consistent size of the ROI of the subsequent string module.
2) Support for external or internal scaling of test results.
Applicable scenarios:
Able to expand or shrink the inspection results to address some edge pixel-level omissions and over inspection.
5. Upgraded data management
Introduction of the Tag concept.
Image identification alias management feature (Tag Management) for marking images to reduce the confusion of unmanageable data sets in large data volume projects.
6. Upgraded annotation class
Adding an annotation management feature.
7. Upgraded model training & validation
Automatic multi-task queuing
Upgraded confusion matrix
Provide more intuitive model evaluation results to guide model iteration
8. Upgraded image list
The image list adds six new elements: defect annotation, inspection results, full image mask, key learning area, no learning area, and ROI.
