Prof. Javad Shafiee

Dec 10, 2022, 11:30am EC4-2101A

Network forgetting is a very common phenomenon whereby networks forget previous training which is not continuously reinforced.  In this talk I will propose a fast and robust video object segmentation method which mitigates the network forgetting that can happen during online updating/retraining because of limited memory.  I will show that a proposed method (I-FRTM) is faster on the DAVIS2016 dataset in comparison with FRTM, the current state of the art. Additionally, I will propose a new test scenario, that accentuates the forgetting challenge, and that demonstrates how the proposed method is more robust against forgetting during update.