Continual Learning setting
pierrefournier752 opened this issue · comments
Hi @MSiam Thank you for sharing the code of your paper, which I found very interesting. I would like to know how to reproduce the results for the Continual learning part. I am able to reproduce the results for the standard few-shot (both in mIOU and BF-mIOU) but I do not know how to do the same for continual learning. Also, from the Fig.6 in your paper (ICCV version) the performance after having learned just 1 additional tasks drops to ~0.1%. Does this mean that after updating the parameters for the first set of 2 novel classes the performance on old classes goes to 0.1? Thanks!!
Ok, I think I have found it in another issue. I will check your answer there and come back if I have further questions.