Dear Organizers,
We trust that you will conduct a comprehensive review of the top prize winners, allowing us to examine their training code, approaches and logs in detail, along with the data that they used, due to the train-test leak.
Thank you!
Posted by: siom.keling @ March 3, 2024, 9:11 a.m.Yes, totally agree. The task of p1 is easy, but the tasks of p2.1 ang p2.2 are hard. It seems that p2.1 task is to use the deepfake data to train a model, tested on the face anti-spoofing data, and the p2.2 task is just the opposite(i.e.,using the face anti-spoofing data to train a model, tested on the deepfake data). In my opinion, the tasks are insteresting, but it seems like the model trained on deepfake data(or face anti-spoofing data) is challenge to transfer to face anti-spoofing data(or deepfake data).
Posted by: yiyao_scb @ March 3, 2024, 12:50 p.m.