Face Image Animation testing with our own pictures

[ms-ufad]

The face image animation task, when testing with our own pictures and videos, is to put the pictures in the ./demo_ Results/face/ ,
But there was no face change in the video。

[RenYurui]

Hi,
Currently, our model cannot be used for arbitrary face image animation.
One way to improve it is to retrain our model in larger face image dataset.
Meanwhile, removing the background edges from the input instruction may help to improve the performance of the cross-identity animation task.

[RenYurui]

Datasets such as VoxCeleb can be used for this task.

[ms-ufad]

Thank you very much for your reply. Maybe I didn't describe it clearly. Can I input an image and a video to replace the face in the video with the face in the image?

[ms-ufad]

I see this sentence: “Given a source face and a sequence of edge images, our model generates the result video with specific motions.”
Where is the implementation code?

[RenYurui]

Hi, we generate new videos by extracting motions from the driving videos.
I think you may want something like face swapping. However, our model is a tool for face reenactment.
The difference can be found in many papers such as https://arxiv.org/abs/2008.02793

[ms-ufad]

python demo.py
--name=face_checkpoints
--model=face
--attn_layer=2,3
--kernel_size=2=5,3=3
--gpu_id=0
--dataset_mode=face
--dataroot=./dataset/FaceForensics
--results_dir=./demo_results/face
你好，我用自己录制的一段朗读视频，运行上面这段代码后，得到了每帧图像对应的人脸motion草图以及一个视频。
现在的疑问是怎么根据这些motion图像和另一张目标人脸图像，生成一段目标人脸的视频，像根据人体motion图生成对应视频序列那样。

[Zhenggen-deng]

python demo.py
--name=face_checkpoints
--model=face
--attn_layer=2,3
--kernel_size=2=5,3=3
--gpu_id=0
--dataset_mode=face
--dataroot=./dataset/FaceForensics
--results_dir=./demo_results/face
你好，我用自己录制的一段朗读视频，运行上面这段代码后，得到了每帧图像对应的人脸motion草图以及一个视频。
现在的疑问是怎么根据这些motion图像和另一张目标人脸图像，生成一段目标人脸的视频，像根据人体motion图生成对应视频序列那样。

你的问题解决了吗/