Neurocle Overcomes the Limitations of Traditional Transfer Learning with Neuro-T & Neuro-R v2.3

Increasing the Accuracy and Efficiency of Deep Learning Projects

Neurocle (CEO Hongsuk Lee), a company specializing in deep learning-based vision software, has released Neuro-T and Neuro-R version 2.3 which includes a variety of new features.

Neuro-T and Neuro-R are deep learning vision software that even non-experts can use. The intuitive graphical UI (GUI)-based design allows non-experts to take advantage of deep learning vision technology in which machines learn and analyze images on their own. Without additional coding, users can use deep learning vision technology to design automated processes such as defect inspections, medical image analysis, logistics packaging inspections, and more.

▲ Fast Retraining function added in version 2.3

The noteworthy feature of this release is the Fast Retraining feature, which enables rapid maintenance.

Previously, product inspection was often challenged by rapid replacement of products or the addition of defect types. Many customers attempted to solve this issue with transfer learning, a method of adding data to an existing deep learning model for training. In reality, the changes were not clearly reflected due to the constraints of the existing model. However, the Fast Retraining feature is a learning method that leverages the existing architecture and hyperparameters when training data is added or changed, resulting in a better quality of maintenance while also drastically reducing retraining time.

Version 2.3 also includes more ways to improve the accuracy of the model with image pre-processing, specifically the ability to set the Region of Interest (ROI) or mask an image. Other improvements include the option to choose data augmentation, enhanced Labeling Tab UI, keyboard shortcuts, and other tools. These improvements reflected customer feedback from previous versions.

Neurocle CEO Hongsuk Lee said, "Since its early versions, Neurocle has provided customers with convenient model training by using the Auto Deep Learning algorithm to automatically generate models. With this release, we wanted to further maximize the efficiency and economic value of deep learning projects by adding the Fast Retraining function and image pre-processing, considering situations where training data must change. We will continue to attentively listen to our customers and grow through continuous research and development."

Meanwhile, Neurocle has recently successfully registered a number of key technology patents based on its innovative products and has successfully signed contracts with leading conglomerates and university hospitals, posting sales growth of more than 200% annually. In addition, Neurocle has gained both national and international recognition through the ImpaCT-ech Grand Prize from Korean Ministry of Science and ICT, and the Innovators Award from Vision System Design.