something about train #170

New Issue

2021-12-06T08:08:01+01:00

monument-and-sea-all-the-gift commented

2021-12-06 08:08:01 +01:00

(Migrated from github.com)

what's value the id_loss should to achieve so that the simswap have good result,looking forward to your reply，thank you

👍 2

monument-and-sea-all-the-gift commented

2021-12-06 10:08:17 +01:00

(Migrated from github.com)

def forward(self, img_id, img_att, latent_id, latent_att, for_G = False):

the next question about tag(for_G ),is it have some effects？thank you

def forward(self, img_id, img_att, latent_id, latent_att, for_G = False): ================================================ the next question about tag(for_G ),is it have some effects？thank you

👍 1

kyugorithm commented

2022-02-08 16:05:55 +01:00

(Migrated from github.com)

On question #1
The ID loss uses cosine similarity (applied in the face recognition network family a.k.a ArcFace)
This is to increase the similarity between the two images in terms of ID.
If model is learned in a desirable direction by this loss, the generated image reflects the ID of the source image well.

On question #1 The ID loss uses cosine similarity (applied in the face recognition network family a.k.a ArcFace) This is to increase the similarity between the two images in terms of ID. If model is learned in a desirable direction by this loss, the generated image reflects the ID of the source image well.

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: CalvinBackup/neuralchen-SimSwap#170