v1.1 - BIG update

BIG UPDATE:

Please uninstall your gsplat since it has big bugs in version 0.1.7

pip uninstall nerfstudio gsplat
pip install git+https://github.com/nerfstudio-project/gsplat.git
pip install git+https://github.com/ichsan2895/nerfstudio.git

Now, it based Latest Nerfstudio (commit ae6c46c, 9 March 2024). So it supports of this features:

1. Splatfacto-big is supported now. splatfacto-big
2. Flag --pipeline.model.rasterize_mode antialiased in ns-train splatfacto and/or ns-train splatfacto-big Implement new rasterization mode using opacity compensation factor
3. You change the background to alpha channel in RGBA images, so it will be transparent in splatfacto. Add transparency carving to splatfacto
4. Enable coarse-to-fine training by default. It seems increase the result quality. (nerfstudio-project#2984)

Inital release

What is different between original repo and this fork?

1. This fork has ns-process-data splatfacto
2. This fork uses ALL IMAGES for training instead of 0.9*All images in original Nerfstudio. But, eval images is splitted by fraction or interval. Default settings is fraction. Eval images is 0,1 of all images.

Why uses all training images?

It mimics INRIA 3DGS which uses all images for training data.

Will it be overfitting?

Yes, thats okay for me. Overfitting is not problem because we don't generate entirely new scene. But we must maximize the existing scene quality.

You can NOT compare the metrics since you have trained all images included eval images.

Yes it will be biased since eval images already leaks into training dataset. But this repo is intended for END USER which does not care about PSNR, SSIM, etc. They care only the end product.