Vanleya

The control conditions have basically locked in the positions that need to appear. Can we use pattern control, but not specify the specific location?

how to measure face similarity

1

in 4 Experiments: RetinaFace and FaceNet but in appendix: Arcface which one is the right one?

Will adding two noise directly during inference affect the generation effect? As shown in the picture, it feels a bit chaotic 

after 3d reconstruction, we have a mesh or obj file, then get renderedface normals or image? how to get? thank you

Has anyone reproduced this paper? When will the author release the training code？Thanks

仿照了ip-adapter的解耦交叉注意力机制 ip-adapter中是在每个交叉注意力层后又添加了处理图像特征的交叉注意力层 添加的这一层是需要训练的 请问instantID中没有提到这一层的训练，应该如何理解呢？ The decoupled cross-attention mechanism was modeled after the IP-Adapter. In the IP-Adapter, after each cross-attention layer, an additional cross-attention layer that processes image features is added....

有人想一起做这篇论文的复现吗？

5

我们可以一起讨论论文细节尝试复现什么的

when we train Conventional Diffusion branch, we usually generate noise randomly->image added noise(x_t)->other condition->unet predict noise->calculate loss so we want reconstruct x_0 from x_t, then compare face why in the...