nir-vana innovation network

The Romanian company, based in Transylvania, supports businesses in accessing novel solutions based on the most recent advances in Artificial Intelligence and Machine Learning: ● They design customized solutions based on cutting edge technologies ● Their software are easy to use with simple and intuitive user interfaces ● They focus on explainable models and effective visualization tools The Romanian company has developed several products which can be modified and adjusted according to customer needs: Optical Character Recognition (OCR) ● Robust models in the presence of characters with deformations, occlusions, and blurred images ● Stunning accuracy also with non-standards fonts and reflective surfaces, where traditional OCR fails. ● Optical Character Verification (OCV) based on user-defined patterns ● Possibility to refine the model online with annotated images for brilliant performances on the most difficult OCR scenarios Object Detection ● Detect position and orientation of objects in images, train models with few images per class ● Pre-trained out-of-the-box models for generic object detection capable of identifying position and orientation of more than 600 objects in an image ● Robust models recognizing objects in the presence of rotations, deformations and partial occlusions ● Customized models trained using ad-hoc synthetic datasets for domain-specific object detection with superior performances on the most challenging tasks ● Zero-shot object detection allows to detect images of objects never seen in training Anomaly Detection ● Detect anomalies in regular surfaces with regular patterns, such as cracks, defects, discoloration, also when they are unknown a priori ● Pre-trained out-of-the-box models capable of identifying anomalies in regular surfaces with regular patterns, such as cracks, defects, discoloration ● Customized models trained with only few images of flawless cases can successfully identify anomalous examples of unlikely appearance in images coming from specific domains ● Anomalous samples can be provided to the model to better discriminate anomalies and tune the sensitivity and determine the acceptable sample variability ● Anomalies are highlighted in the image with heatmaps to highlight missing or extra elements and support model explainability. The company is looking for partners via commercial or outsourcing agreements.