Summer job – object detection on minimal annotated data 


VOCA is an AI technology development company providing tools for digitalization, remote control and autonomy, targeted uncontrolled and harsh environments. Located in Norway, VOCA serves an international market. Our team members have extensive experience ranging from technology development within AI, computer vision and machine control to strategy, customer relations and international business development. VOCA is always looking for new ways to improve and therefore would like students to work on making state of the art technologies industry-ready. 


Deep learning is often done in a fully supervised fashion, with manually annotated data. This is time-consuming and expensive. Therefore VOCA would like students to look into ways to reduce its current need for annotated data. There are several promising technologies within the realm of semi-supervised learning that allows for training models with minimal amounts of annotated data.  

The students will be given a particular object detection task and a limited amount of annotated data, and will then try to match or overcome the performance of VOCAs own model. Transfer learning is also relevant as the students will be given datasets from different domains.  

The students will get to work on solving a highly applicable industry type problem while using real-world data. The goal is for the student to create an overview of relevant methods and approaches, for then to test and compare them in terms of industrial needs. 

Research questions 

  • How does the ratio of annotated and un-annotated data influence the performance of ML models? 
  • In what sense can un-supervised methods be used to improve performance and robustness, and speed up transfer learning of ML models.  


  • Should have experience with programming. 
  • Should be familiar with basic concepts of AI and have experience from previous AI projects. 

Similar work / Links  


Henrik Brådland, Machine Learning Engineer  

+47 482 35 984,