Computer Vision News - July 2022
BEST OF CVPR 10 CVPR Poster Presentation comparable with a fully supervised semantic segmentation model , even with our low annotation costs. This model has the potential to be applied to many real- world applications, and I think that it’s a very valuable contribution.” Fabio and Dario aremissing their co-author Antonio Tavera , who cannot be here to present today. Fabio recalls a special eureka moment they all had together, and without which we may not be speaking: “We had been in the lab all day, and it was a Friday night before the conference deadline, and we were in a restaurant with Antonio, still talking about our model. That’s when we had the idea of the final model that got us these last performances. I remember telling Dario we can try that tomorrow, and it was a very last-minute tweak!” The novelty of this work and its application in the real world are just two things that likely helped it to be chosen from so many an old task, but this is the first time both tasks have been combined. The team hopes it will become a new field. To train incrementally over time, the model uses the knowledge of previous timestamps to maintain the knowledge of the classes that it has learned earlier to avoid forgetting them. The team uses techniques like segmentation loss , which allow the model to predict a label for each single pixel, and in this specific scenario, there is a common solution that averages the model’s features to be trained at the image-level label. “We use a technique called CAM, or Class Activation Maps , and make an average of the scores on the images to train them on the image-level labels,” Fabio explains. “We use this at the image level to extract a localization for the new classes, which helps us to train a segmentation model.” Dario adds: “We’ve achieved results
Made with FlippingBook
RkJQdWJsaXNoZXIy NTc3NzU=