det-size and max face similarity threshold (language confusion) #100
-
|
Hi everyone det-size: So this determines the size of the "box", in which a face gets detected, right? It will detect faces in a 640x640-box. If the whole face isn't inside those 640x640, it won't get detected/swapped. If I have a 1280x720 video, close-ups won't get swapped. Assuming I use the same video, but with a resolution 1920x1080, it will detect/swap even less, right? So what would be the use case for a smaller detection window of 320x320? And if I'd want to have the close-ups swapped too, I'd need a bigger detection window, not smaller? Or am I misunderstanding something? max face similarity threshold (I always get confused with this word): Everything above that value will get swapped, right? So shouldn't it say min face similarity, since the face has to have a minimum calculated face similarity of 0.65 (default), to get swapped? I'm confused. Thanks in advance for your help. |
Beta Was this translation helpful? Give feedback.
Replies: 2 comments
-
|
I'm no english native speaker either, this is fine with me (and seeing how you wrote 'Englisch' it might even be the same country). det-size: the internals of the insightface model are unknown and treated as a black box. By experimenting I and others found out that the default 640x640 size doesn't detect LARGE closeups of faces. The checkbox just toggles between the standard size of 640x640 and the smaller size 320x320. So it's actually the other way around, it needs to be smaller to detect larger closeups! However a smaller size also seems to neglect finer details in the face, be aware of that. max_similarity: |
Beta Was this translation helpful? Give feedback.
-
|
lol the classic typo. possibly, yes. or at least "one of those" countries ;-) aaalright, now it makes sense. funny how my brain turns everything around 180° lately :D |
Beta Was this translation helpful? Give feedback.
I'm no english native speaker either, this is fine with me (and seeing how you wrote 'Englisch' it might even be the same country).
det-size: the internals of the insightface model are unknown and treated as a black box. By experimenting I and others found out that the default 640x640 size doesn't detect LARGE closeups of faces. The checkbox just toggles between the standard size of 640x640 and the smaller size 320x320. So it's actually the other way around, it needs to be smaller to detect larger closeups! However a smaller size also seems to neglect finer details in the face, be aware of that.
max_similarity:
no, it's again the other way around 😏
The wording is misleading but I didn't n…