Docs (#18)

* update doc title

* update to pip install

* cleanup docs

docs/source/api.rst

-.. _api:
-
-API
-============
-
-.. automodule:: nerfacc
-   :members:
-   :show-inheritance:
-

docs/source/apis/occupancy_field.rst

+OccupancyField
+===================================
+
+.. currentmodule:: nerfacc
+
+.. autoclass:: OccupancyField
+    :members:
+    :show-inheritance:
\ No newline at end of file

docs/source/apis/volumetric_marching.rst

+Volumetric Ray Marching
+=========================
+
+.. currentmodule:: nerfacc
+
+.. autofunction:: ray_aabb_intersect
+.. autofunction:: volumetric_marching
\ No newline at end of file

docs/source/apis/volumetric_rendering.rst

+Volumetric Rendering
+======================
+
+.. currentmodule:: nerfacc
+
+.. autofunction:: volumetric_rendering_steps
+.. autofunction:: volumetric_rendering_weights
+.. autofunction:: volumetric_rendering_accumulate
\ No newline at end of file

docs/source/examples/ngp.rst

+Instant-NGP
+====================
+
+See code at our github repository: https://github.com/KAIR-BAIR/nerfacc/examples/
+
+Benchmarks
+------------
+
+We trained on NeRF-Synthetic trainval set using TITAN RTX, and evaluated on testset.
+Note Instant-NGP's results are taken from the paper, which is trained on a Nvidia 3090,
+with random background trick using alpha channel.
+
++----------------------+----------+----------+------------+-------+--------+
+|                      | Lego     | Mic      | Materials  |Chair  |Hotdog  |
+|                      |          |          |            |       |        |
++======================+==========+==========+============+=======+========+
+| Paper (PSNR: 5min)   | 36.39    | 36.22    |  29.78     |  35.00| 37.40  |
++----------------------+----------+----------+------------+-------+--------+
+| Ours  (PSNR)         | 36.61    | 37.45    | 30.15      | 36.06 | 38.17  |
++----------------------+----------+----------+------------+-------+--------+
+| Ours  (Training time)| 300s     | 272s     | 258s       | 331s  | 287s   |
++----------------------+----------+----------+------------+-------+--------+

docs/source/index.rst

+:github_url: https://github.com/KAIR-BAIR/nerfacc
 
-.. image:: _static/images/logo.png
-  :width: 400
-  :align: center
-  :alt: nerfacc
+NeRFacc Documentation
+===================================
 
-NeRF Acceleration
------------------
-
-This is a fast differentiable volume rendering toolbox as a PyTorch extension for NeRF.
-
-Check out the :doc:`usage` section for further information, including
-how to :ref:`installation` the project.
+NeRFacc is a PyTorch NeRF acceleration toolbox for both training and inference. 
 
 .. note::
 
    This project is under active development.
 
-Contents
---------
+Installation:
+-------------
+
+.. code-block:: console
+
+   $ pip install nerfacc
+
+
+.. toctree::
+   :glob:
+   :maxdepth: 1
+   :caption: Example Usages
+
+   examples/*
+
+
+.. toctree::
+   :glob:
+   :maxdepth: 1
+   :caption: Python API
+
+   apis/*
 
 .. toctree::
+   :maxdepth: 1
+   :caption: Projects
 
-   usage
-   api
\ No newline at end of file
+   NeRFactory <https://plenoptix-nerfactory.readthedocs-hosted.com/>

docs/source/usage.rst

-Usage
-=====
-
-.. _installation:
-
-Installation
-------------
-
-To use nerfacc, first install it using pip:
-
-.. code-block:: console
-
-   (.venv) $ pip install git+https://github.com/liruilong940607/nerfacc
-
-Example of use
-----------------
-
-.. code-block:: python
-
-    from typing import Callable, List, Union
-    from torch import Tensor
-    import torch
-    import torch.nn.function as F
-
-    from nerfacc import OccupancyField, volumetric_rendering
-
-    # setup the scene bounding box.
-    scene_aabb = torch.tensor([-1.5, -1.5, -1.5, 1.5, 1.5, 1.5]).cuda()
-
-    # setup the scene radiance field. Assume you have a NeRF model and 
-    # it has following functions:
-    # - query_density(): {x} -> {density} 
-    # - forward(): {x, dirs} -> {rgb, density}
-    radiance_field = ...
-
-    # setup some rendering settings
-    render_n_samples = 1024
-    render_bkgd = torch.ones(3).cuda()
-    render_step_size = (
-        (scene_aabb[3:] - scene_aabb[:3]).max() * math.sqrt(3) / render_n_samples
-    )
-    
-    # setup occupancy field with eval function
-    def occ_eval_fn(x: torch.Tensor) -> torch.Tensor:
-        """Evaluate occupancy given positions.
-
-        Args:
-            x: positions with shape (N, 3).
-        Returns:
-            occupancy values with shape (N, 1). 
-        """
-        density_after_activation = radiance_field.query_density(x)
-        occupancy = density_after_activation * render_step_size
-        return occupancy
-    occ_field = OccupancyField(occ_eval_fn=occ_eval_fn, aabb=aabb, resolution=128)
-
-    # training
-    for step in range(10_000):
-        # generate rays from data and the gt pixel color
-        rays = ...
-        pixels = ...
-
-        # update occupancy grid
-        occ_field.every_n_step(step)        
-
-        # rendering
-        (
-            accumulated_color,
-            accumulated_depth,
-            accumulated_weight,
-            _,
-        ) = volumetric_rendering(
-            query_fn=radiance_field.forward,  # {x, dir} -> {rgb, density}
-            rays_o=rays.origins,
-            rays_d=rays.viewdirs,
-            scene_aabb=aabb,
-            scene_occ_binary=occupancy_field.occ_grid_binary,
-            scene_resolution=occupancy_field.resolution,
-            render_bkgd=render_bkgd,
-            render_n_samples=render_n_samples,
-            # other kwargs for `query_fn`
-            ...,
-        )
-
-        # compute loss
-        loss = F.mse_loss(accumulated_color, pixels)