AI葵

Can you share the tensorboard logs for the 2 trainings? And the final checkpoint

Hi, I have never encountered this kind of error in my experiment (maybe nan occurs but no error appeared) If you want to change density grid, you can do it...

I don't have MacOS so I cannot test; if you have MacOS, you can build the binary from the unity project I provide, and if it works, you are welcome...

Thanks a lot for your contribution! I have updated the release and included your binary. If any MacOS user has problem, feel free to report it here.

I recommend you to build it by yourselves... Download the unityproject.zip [here](https://github.com/kwea123/VTuber_Unity/releases) and open it in unity then build. It should be fast.

I use 2019.01 but I think any version (if unity doesn't complain) should be fine. There is no sophisticated assets used.

Change this parameter to 48: https://github.com/Lotayou/Face-Renovation/blob/b5f5e9e86ec8ebdebcf64d22d806085fead80b8b/options/config_hifacegan.py#L59 It controls the input channel, 48 means 48x16=768 channels. https://github.com/Lotayou/Face-Renovation/blob/b5f5e9e86ec8ebdebcf64d22d806085fead80b8b/models/networks/generator.py#L195-L206 The training will work in this case, but I don't know if the result...

If you install `opencv-python`, there is a more concise way, add the following lines after `spericalCoord = self._calcSphericaltoGnomonic(convertedScreenCoord)` ``` sphericalCoord = sphericalCoord.reshape(self.height, self.width, 2).astype(np.float32) % 1 out = cv2.remap(frame, sphericalCoord[...,...

[My implementation](https://github.com/kwea123/nerf_pl) in pytorch-lightning, faster, less VRAM required, and multi-gpu training is supported! Update: I also provide code to reconstruct **colored** mesh. Update: Tutorial, and many other features on Unity...

Alternatively, I discovered [BlenderProc](https://github.com/DLR-RM/BlenderProc) that allows you to generate images with set camera parameters as well. It requires more time to learn how to use it though.