Matthias Humt

Hi @Zhengxinyang, thank you for the great work! Do you actually compute one FID score for each view and then average across all views (i.e. one FID score for each...

Another solution, which also seems to work and is quite a bit faster than using raytracing, is to do the following before rendering the mesh: ```python mesh_inv = mesh.copy() mesh_inv.invert()...

[It does work](https://github.com/hummat/bproc-pubvis/blob/df158eac2b8f58164eda8e77d980041350e4a248/_config.yaml#L14) for me [here](https://hummat.github.io/bproc-pubvis/).

I implemented it primarily to reproduce results from the original [Variational Diffusion Models](https://arxiv.org/abs/2107.00630) paper and the [Locally Attentional SDF Diffusion for Controllable 3D Shape Generation](https://arxiv.org/abs/2305.04461) paper. I didn't use it...

@yiyixuxu

Thanks for the great experiment @vladmandic! To be honest I'm surprised that it works at all as the current implementation follows the VDM paper where models are expected to be...

> question: this > > https://github.com/lucidrains/vector-quantize-pytorch/blob/1bce1c3b80296f64612f808942460c3a955dec3f/vector_quantize_pytorch/lookup_free_quantization.py#L244 > > disables autocast while the recent #145 enables it by default. Should we enable it only if it is already enabled? Sorry for...

@lucidrains thanks for the quick reply. I'm on `1.17.1` from pip but even this master branch has it, right? https://github.com/lucidrains/vector-quantize-pytorch/blob/c302cf3282161e81ebcf77a627b6d0e8bf34b069/vector_quantize_pytorch/finite_scalar_quantization.py#L162

@lucidrains aha, yes, in 1.17.3 it's fine :)

Same problem here. Neither few nor single account QR codes work for me.