Computer Vision News 24 Deep Learning for the Eyes space. In the next step, methods to pull the two representation spaces of the two modalities closer, Lehan used two methods, class prototype matching and class similarity alignment. The key contribution was the ability to bridge the gap between different modalities without paired data. The distillation approach involved training the OCT student model with the knowledge from the fundus teacher model and finally allowed single-modality use during inference. Last but definitely not least, in order to submit (and get accepted) to MICCAI, writing up the entire process proved a bit trickier than anticipated. Finding the right phrases to explain the pipeline needed some more input from her supervisor, who stayed up late together with Lehan before the deadline. This also showed the importance of finding a supervisor that matches one's energy. I wish Lehan the best for the upcoming years of her PhD, and am looking forward to more fascinating publications from her!
RkJQdWJsaXNoZXIy NTc3NzU=