View in #questions-forum on Slack
@Niloufar_Abaei: Hi all,
I’m working on a face recognition project in surveillance cameras. In real world, some new faces could be captured, which are not in the dataset in training phase. I mean the whole data isn’t available in advance and is presented over time. So, I need to retrain the model while capturing new data, while complete retraining isn’t feasible. Hence, I believe CL could be a great option. So far, I haven’t found any informative paper on CL face recognition. Can someone point me toward any research papers or implementation code?
Any help would be appreciated!
@vlomonaco: Check out this recent paper that may be related: https://arxiv.org/abs/2004.07941
We did a few exps in the past as well:
And this book chapter: https://link.springer.com/chapter/10.1007/978-3-030-33904-3_1
Alberto Del Bimbo has done a lot of work in this area, check his works!
arXiv.org: Continual Learning for Anomaly Detection in Surveillance Videos
SpringerLink: Comparing Incremental Learning Strategies for Convolutional Neural Net
@Niloufar_Abaei: Thank you so much. I’ll go through them
@Federico_Pernici: Hi @Niloufar_Abaei, not strictly Continual Learning (fixed representation) but this paper may be related to what you asking for:
Another very interesting paper is
It changes the internal representation but it is not strictly online (it requires multiple passes over the video sequence).
@Niloufar_Abaei: Thank you @Federico_Pernici. So related!
BTW, I have read many papers including those ones you suggested to me and now the way this model works is crystal clear. But I’m not very confident about the implementation part. I do appreciate it if you share any sample code that gives me more clues.