We are proudly releasing a couple of new Artificial Intelligence microservices! Students, collaborators and even profs are contributing to extending the list of microservices on our Kubernetes cluster. Check them out here: https://icosys.ch/icoservices. All microservices are accessible through REST API following the openapi v3 specification (OAS3).
- Flip-ML. The objective of this microservice is to recognise the orientation of printed document images, typically scanned using multi-function printers. This service can recognise orientations of 0° and 180°. The deep Convolutional Neural Network used for this service was trained on data from the RVL-CDIP dataset. [STUDENT-PROJECT]
- NSFW image detector. The objective of this deep learning microservice is to detect if an image is safe or not-safe-for-work (NSFW). The service recognizes 2 main categories: safe and nsfw. In each of these categories, sub-categories are also recognized. For the category nsfw, the following sub-categories are available: suggestive, nudity and porn, associated to increasing degrees of offensive sexual contents. For the category safe, the system can distinguish between general for images such as landscape, objects or animals and person when a person (or a body part) is present in the image. Finally, safe or nsfw cartoons are also detected by the service, meaning there is a sub-category cartoon for both categories. This microservice employs a large deep neural network that was obtained with a transfer learning strategy from the pre-trained network MobileNetV2. This model was fine-tuned with 4’000 images per sub-category.
- Translation en-fr. The objective of this microservice is to showcase a translation system trained using Marian NMT and a recipe of datasets assembled by us. The model belongs to the transformers family and gives performances that compare favourably well using a BLEU metric against leading systems. [STUDENT-PROJECT]