zhuhaozh

Did this repo can reproduce the comparable results with original paper?

1

Hi, Thanks for sharing your code. I've downloaded this repo and tried to run the demo.m but encountered the following errors.  Then I tried to implements it by myself...

Is the mutual information estimate is wrong when change the data dimension?

8

Hi, I changed the data dimension from 1 to another shape for example, 3*28*28, and modified the model similar to offered in MINE appendix. However, I found that estimated mutual...

decord read failed, but ffmpeg works fine

4

VideoReader (xxx.mp4) reports the error: > decord._ffi.base.DECORDError: [10:17:57] /github/workspace/src/video/video_reader.cc:151: Check failed: st_nb >= 0 (-1128613112 vs. 0) ERROR cannot find video stream with wanted index: -1 I checked the video...

raw audio in your data

1

Hi, Thanks for your nice work. Would it possible for you to release the raw audio of corresponded key point?

Weird results using pretrained model

8

Using the inference.py, got some weird results on my images.  

Questions about kp2gaussian and gaussian2kp

1

Hi, Thanks for your interesting work! I have some questions about the kp2gaussian and gaussian2kp in keypoint_detector.py 1. Are these functions invertible? I use a pretrained hourglass network to extract...

Thank you for releasing the code so quickly

2

It is truly an interesting work. Are there any plans to provide colab scripts or huggingface demo?

Use the pretrained model ,but got the wrong test result

1

My PC environment is python3.6+pytorch0.40. After I changed and only changed the str(01) to "01" in IdentityEncoder.py, this code(test_all.py) can run well, with some deprecated UserWarning,But the output is far...

Hi! Although currently there are 7 agents are provided, How can i add more characters? Where should I put them to and which format should be used?

代码实现细节与原论文不一致？

3

Hi, @pengbo-learn 我最近也有打算复现这篇论文，但发现复现的结果有点问题。在您这个repo下，我发现您与原论文中的一些实现似乎不太一样： 原论文中是使用color, position, normal 先渲染图片后，使用poisson对非脸部区域做了一个补全，然后三个渲染的结果concat在一块后，使用pdf transfer，得到relighting后的图片。 但您这里的实现只是用了 3d mesh上的点的信息做 pdf tranfer， 输出color信息后再渲染到原图像上。我尝试使用3ddfa来代替您的使用的prnet，但发现结果并不太满意，只能够做到迁移整体的全局颜色，而无法做到局部的光照改变。 我也尝试将实现的严格按照论文中的说法实现，但效果还是与之前一样。 不知道您对这样的现象有没有什么建议？下图我是自己复现的结果