kiui

In fact `aabb_scale` has no use in this codebase, it's replaced by `--bound` as stated in the readme. You may use the same `aabb_scale` as `bound` for the original instant-ngp...

@Zaki0207 Hi, does the dataset contains background, or only the object with alpha channel? You could try to use the GUI mode to debug. Also, could you post the full...

@Zaki0207 Could you try to use a larger bound, like 4 or 8? Another option is to try to add `--bg_radius 32`.

@Xallt Hi, don't forget to reinstall the extension after updating the code? I got the following results with the latest commit: ``` In [7]: enc = GridEncoder(input_dim=3).cuda() ...: enc(torch.rand(3).cuda()) Out[7]:...

@AIBluefisher Hi, thanks for your information! Do you refer to [this line](https://github.com/ashawkey/torch-ngp/blob/main/gridencoder/grid.py#L99)? I'm just following [this part](https://github.com/NVlabs/instant-ngp/blob/a463dae3bee74d42a35065148d138bf6ff562895/src/testbed.cu#L2237) from instant-ngp. Could you provide more details, e.g., the full command you use...

A possible reason is that instant-ngp loads all train/val/test data to train (see [here](https://github.com/NVlabs/instant-ngp/blob/7b37eee0aa9ea31362d2a5a799e007a731de385c/src/testbed_nerf.cu#L2211)) for nerf_synthetic dataset?

@chenerg Hi, it's from instant-ngp's implementation, please check their code and paper.

@ug-kim Hi, the code of DMTet is majorly adapted from [nvdiffrec](https://github.com/NVlabs/nvdiffrec). You may check their paper for a detailed explanation. In summary, we still use the 3D hashgrid from NeRF...

Hi, I think the easiest way is to convert the checkpoint to diffuser-compatible format, with tools like [this](https://huggingface.co/spaces/diffusers/sd-to-diffusers). Then you could use the `--hf_key` option to load custom checkpoints. For...

Hi, unfortunately the quality is majorly dependent on the quality of 4 multi-view images. Hopefully we can use better multi-view diffusion models like this: https://yashkant.github.io/spad/