search

grispeut / Feature-Alignment

object detection, adversarial robustness, ICIP2021
17 stars 1 forks source link

Can this method really improve the map value on clean data? (take no account of robustness) #1

Open huhengtong opened 3 years ago

huhengtong commented 3 years ago

I am very curious about if your method (or other adversarial training methods) can truly improve the map value of object detection task on clean data like MSCOCO?

grispeut commented 3 years ago

emmmm, the approach can only improve map than existing adversarial training methods. 

---Original--- From: "huhengtong"<notifications@github.com> Date: Fri, Jan 22, 2021 16:51 PM To: "grispeut/Feature-Alignment"<Feature-Alignment@noreply.github.com>; Cc: "Subscribed"<subscribed@noreply.github.com>; Subject: [grispeut/Feature-Alignment] Can this method really improve the map value on clean data? (take no account of robustness) (#1)

I am very curious about if your method (or other adversarial training methods) can truly improve the map value of object detection task on clean data like MSCOCO?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or unsubscribe.