Faceswap Forum

i just saw the message :
05/10/2023 23:43:41 INFO Sorting by identity similarity...
05/10/2023 23:43:41 INFO VGG Face2 Embeddings are being written to the image header. Sorting by this method will be quicker next time

actually how to use that ?> thank you

VGGFace2 embeddings are used for the face filter (I generally don't use that, but it is available) and for "Sort by Face".

For the latter, storing the embeddings at extract time helps save time at the sort stage, as the information is held in the PNG headers, which is a lot quicker than having to calculate them whilst sorting.

If you haven't stored the embeddings at extract time, then they are stored when you sort by face, so if you sort by face again, it is a lot quicker!

is there any setting can set batchsize of components and extended ?

05/11/2023 19:18:14 INFO Initializing Components (Mask)...
05/11/2023 19:18:14 INFO Initialized Components (Mask) with batchsize of 1
05/11/2023 19:18:14 INFO Initializing Extended (Mask)...
05/11/2023 19:18:14 INFO Initialized Extended (Mask) with batchsize of 1

Before i spent like 3hours extract for 40mins video with _80000png
so i just test servral setting of extract batchsize , let see what different is .....

No.

Extended and Components are not Neural Network masks. They are generated from the landmarks, so the premise of 'batch size' is meaningless for them. It just outputs that information, as it is standard plugin output.