deploy: f92fd3f

main/_sources/examples/colmap.rst.txt

@@ -15,7 +15,7 @@ Simply run the script under `examples/`:
 
 .. code-block:: bash
 
-    python simple_trainer.py \
+    CUDA_VISIBLE_DEVICES=0 python simple_trainer.py \
         --data_dir data/360_v2/garden/ --data_factor 4 \
         --result_dir ./results/garden
 

main/_sources/examples/large_scale.rst.txt

@@ -35,7 +35,7 @@ The code for this example can be found under `examples/`:
 .. code-block:: bash
 
     # First train a 3DGS model
-    python simple_trainer.py \
+    CUDA_VISIBLE_DEVICES=0 python simple_trainer.py \
         --data_dir data/360_v2/garden/ --data_factor 4 \
         --result_dir ./results/garden
 

main/_sources/index.rst.txt

@@ -13,13 +13,21 @@ Overview
 Real-Time Rendering of Radiance Fields" :cite:p:`kerbl3Dgaussians`, but we've made *gsplat* even 
 faster, more memory efficient, and with a growing list of new features!
 
-* *gsplat* is developed with efficiency in mind. Comparing to the `official implementation <https://github.com/graphdeco-inria/gaussian-splatting>`_, *gsplat* enables up to **4x less training memory footprint**, and up to **15% less training time** on Mip-NeRF 360 captures, and potential more on larger scenes. See :doc:`tests/eval` for details.
+* *gsplat* is developed with efficiency in mind. Comparing to the `official implementation <https://github.com/graphdeco-inria/gaussian-splatting>`_, 
+  *gsplat* enables up to **4x less training memory footprint**, and up to **15% less training time** on Mip-NeRF 360 captures, and potential more on larger scenes. See :doc:`tests/eval` for details.
 
-* *gsplat* is designed to **support extremely large scene rendering**, which is magnitudes faster than the official CUDA backend `diff-gaussian-rasterization <https://github.com/graphdeco-inria/diff-gaussian-rasterization>`_. See :doc:`examples/large_scale` for an example.
+* *gsplat* is designed to **support extremely large scene rendering**, which is magnitudes 
+  faster than the official CUDA backend `diff-gaussian-rasterization <https://github.com/graphdeco-inria/diff-gaussian-rasterization>`_. See :doc:`examples/large_scale` for an example.
 
-* *gsplat* offers many extra features, including **batch rasterization**,  **N-D feature rendering**, **depth rendering**, **sparse gradient** etc. See :doc:`apis/rasterization` for details.
+* *gsplat* offers many extra features, including **batch rasterization**,  
+  **N-D feature rendering**, **depth rendering**, **sparse gradient**, 
+  **multi-GPU distributed rasterization**
+  etc. See :doc:`apis/rasterization` for details.
 
-* *gsplat* is equipped with the **latest and greatest** 3D Gaussian Splatting techniques, including `absgrad <https://ty424.github.io/AbsGS.github.io/>`_, `anti-aliasing <https://niujinshuchong.github.io/mip-splatting/>`_ etc. And more to come!
+* *gsplat* is equipped with the **latest and greatest** 3D Gaussian Splatting techniques, 
+  including `absgrad <https://ty424.github.io/AbsGS.github.io/>`_, 
+  `anti-aliasing <https://niujinshuchong.github.io/mip-splatting/>`_,
+  `3DGS-MCMC <https://ubc-vision.github.io/3dgs-mcmc/>`_ etc. And more to come!
 
 
 .. raw:: html

main/_sources/tests/eval.rst.txt

@@ -3,17 +3,19 @@ Evaluation
 
 .. table:: Performance on `Mip-NeRF 360 Captures <https://jonbarron.info/mipnerf360/>`_ (Averaged Over 7 Scenes)
 
-    +------------+-------+-------+-------+------------------+------------+
-    |            | PSNR  | SSIM  | LPIPS | Train Mem        | Train Time |
-    +============+=======+=======+=======+==================+============+
-    | inria-7k   | 27.23 | 0.829 | 0.204 | 7.7 GB           | 6m05s      |
-    +------------+-------+-------+-------+------------------+------------+
-    | gsplat-7k  | 27.21 | 0.831 | 0.202 | **4.3GB**        | **5m35s**  |
-    +------------+-------+-------+-------+------------------+------------+
-    | inria-30k  | 28.95 | 0.870 | 0.138 | 9.0 GB           | 37m13s     |
-    +------------+-------+-------+-------+------------------+------------+
-    | gsplat-30k | 28.95 | 0.870 | 0.135 | **5.7 GB**       | **35m49s** |
-    +------------+-------+-------+-------+------------------+------------+
+    +---------------------+-------+-------+-------+------------------+------------+
+    |                     | PSNR  | SSIM  | LPIPS | Train Mem        | Train Time |
+    +=====================+=======+=======+=======+==================+============+
+    | inria-7k            | 27.23 | 0.829 | 0.204 | 7.7 GB           | 6m05s      |
+    +---------------------+-------+-------+-------+------------------+------------+
+    | gsplat-7k           | 27.21 | 0.831 | 0.202 | **4.3GB**        | **5m35s**  |
+    +---------------------+-------+-------+-------+------------------+------------+
+    | inria-30k           | 28.95 | 0.870 | 0.138 | 9.0 GB           | 37m13s     |
+    +---------------------+-------+-------+-------+------------------+------------+
+    | gsplat-30k (1 GPU)  | 28.95 | 0.870 | 0.135 | **5.7 GB**       | **35m49s** |
+    +---------------------+-------+-------+-------+------------------+------------+
+    | gsplat-30k (4 GPUs) | 28.91 | 0.871 | 0.135 | **2.0 GB**       | **11m28s** |
+    +---------------------+-------+-------+-------+------------------+------------+
 
 This repo comes with a standalone script (:code:`examples/simple_trainer.py`) that reproduces 
 the `Gaussian Splatting <https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/>`_ with
@@ -131,7 +133,7 @@ is different from what's reported in the original paper that uses
 :code:`from lpipsPyTorch import lpips`.
 
 The evaluation of `gsplat-X` can be reproduced with the command 
-:code:`cd examples; bash benchmark.sh` 
+:code:`cd examples; bash benchmarks/basic.sh` 
 within the gsplat repo (commit 6acdce4). 
 
 The evaluation of `inria-X` can be 

main/examples/colmap.html

@@ -324,7 +324,7 @@ <h1>Fit a COLMAP Capture<a class="headerlink" href="#fit-a-colmap-capture" title
 able to reproduce the metrics in the paper, with much better training speed and memory
 footprint. See <a class="reference internal" href="../tests/eval.html"><span class="doc">Evaluation</span></a> for more details on the comparision.</p>
 <p>Simply run the script under <cite>examples/</cite>:</p>
-<div class="highlight-bash notranslate"><div class="highlight"><pre><span></span>python<span class="w"> </span>simple_trainer.py<span class="w"> </span><span class="se">\</span>
+<div class="highlight-bash notranslate"><div class="highlight"><pre><span></span><span class="nv">CUDA_VISIBLE_DEVICES</span><span class="o">=</span><span class="m">0</span><span class="w"> </span>python<span class="w"> </span>simple_trainer.py<span class="w"> </span><span class="se">\</span>
 <span class="w">    </span>--data_dir<span class="w"> </span>data/360_v2/garden/<span class="w"> </span>--data_factor<span class="w"> </span><span class="m">4</span><span class="w"> </span><span class="se">\</span>
 <span class="w">    </span>--result_dir<span class="w"> </span>./results/garden
 </pre></div>

main/examples/large_scale.html

@@ -333,7 +333,7 @@ <h1>Render a Large Scene<a class="headerlink" href="#render-a-large-scene" title
 switch to low resolution if the rendering is slow.</p>
 <p>The code for this example can be found under <cite>examples/</cite>:</p>
 <div class="highlight-bash notranslate"><div class="highlight"><pre><span></span><span class="c1"># First train a 3DGS model</span>
-python<span class="w"> </span>simple_trainer.py<span class="w"> </span><span class="se">\</span>
+<span class="nv">CUDA_VISIBLE_DEVICES</span><span class="o">=</span><span class="m">0</span><span class="w"> </span>python<span class="w"> </span>simple_trainer.py<span class="w"> </span><span class="se">\</span>
 <span class="w">    </span>--data_dir<span class="w"> </span>data/360_v2/garden/<span class="w"> </span>--data_factor<span class="w"> </span><span class="m">4</span><span class="w"> </span><span class="se">\</span>
 <span class="w">    </span>--result_dir<span class="w"> </span>./results/garden
 

main/index.html

@@ -326,10 +326,18 @@ <h2>Overview<a class="headerlink" href="#overview" title="Link to this heading">
 Real-Time Rendering of Radiance Fields” <span id="id1">[<a class="reference internal" href="#id3" title="Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering. ACM Transactions on Graphics, July 2023. URL: https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/.">1</a>]</span>, but we’ve made <em>gsplat</em> even
 faster, more memory efficient, and with a growing list of new features!</p>
 <ul class="simple">
-<li><p><em>gsplat</em> is developed with efficiency in mind. Comparing to the <a class="reference external" href="https://github.com/graphdeco-inria/gaussian-splatting">official implementation</a>, <em>gsplat</em> enables up to <strong>4x less training memory footprint</strong>, and up to <strong>15% less training time</strong> on Mip-NeRF 360 captures, and potential more on larger scenes. See <a class="reference internal" href="tests/eval.html"><span class="doc">Evaluation</span></a> for details.</p></li>
-<li><p><em>gsplat</em> is designed to <strong>support extremely large scene rendering</strong>, which is magnitudes faster than the official CUDA backend <a class="reference external" href="https://github.com/graphdeco-inria/diff-gaussian-rasterization">diff-gaussian-rasterization</a>. See <a class="reference internal" href="examples/large_scale.html"><span class="doc">Render a Large Scene</span></a> for an example.</p></li>
-<li><p><em>gsplat</em> offers many extra features, including <strong>batch rasterization</strong>,  <strong>N-D feature rendering</strong>, <strong>depth rendering</strong>, <strong>sparse gradient</strong> etc. See <a class="reference internal" href="apis/rasterization.html"><span class="doc">Rasterization</span></a> for details.</p></li>
-<li><p><em>gsplat</em> is equipped with the <strong>latest and greatest</strong> 3D Gaussian Splatting techniques, including <a class="reference external" href="https://ty424.github.io/AbsGS.github.io/">absgrad</a>, <a class="reference external" href="https://niujinshuchong.github.io/mip-splatting/">anti-aliasing</a> etc. And more to come!</p></li>
+<li><p><em>gsplat</em> is developed with efficiency in mind. Comparing to the <a class="reference external" href="https://github.com/graphdeco-inria/gaussian-splatting">official implementation</a>,
+<em>gsplat</em> enables up to <strong>4x less training memory footprint</strong>, and up to <strong>15% less training time</strong> on Mip-NeRF 360 captures, and potential more on larger scenes. See <a class="reference internal" href="tests/eval.html"><span class="doc">Evaluation</span></a> for details.</p></li>
+<li><p><em>gsplat</em> is designed to <strong>support extremely large scene rendering</strong>, which is magnitudes
+faster than the official CUDA backend <a class="reference external" href="https://github.com/graphdeco-inria/diff-gaussian-rasterization">diff-gaussian-rasterization</a>. See <a class="reference internal" href="examples/large_scale.html"><span class="doc">Render a Large Scene</span></a> for an example.</p></li>
+<li><p><em>gsplat</em> offers many extra features, including <strong>batch rasterization</strong>,
+<strong>N-D feature rendering</strong>, <strong>depth rendering</strong>, <strong>sparse gradient</strong>,
+<strong>multi-GPU distributed rasterization</strong>
+etc. See <a class="reference internal" href="apis/rasterization.html"><span class="doc">Rasterization</span></a> for details.</p></li>
+<li><p><em>gsplat</em> is equipped with the <strong>latest and greatest</strong> 3D Gaussian Splatting techniques,
+including <a class="reference external" href="https://ty424.github.io/AbsGS.github.io/">absgrad</a>,
+<a class="reference external" href="https://niujinshuchong.github.io/mip-splatting/">anti-aliasing</a>,
+<a class="reference external" href="https://ubc-vision.github.io/3dgs-mcmc/">3DGS-MCMC</a> etc. And more to come!</p></li>
 </ul>
 <div style="position:relative; padding-bottom:56.25%; height:0; width:100%">
    <iframe style="position:absolute; top:0; left:0; width:100%; height:100%" src="https://www.youtube.com/embed/d5vOUYm5k34?si=fQBdwPgoi0FZOSyD" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" referrerpolicy="strict-origin-when-cross-origin" allowfullscreen></iframe>

main/tests/eval.html

@@ -351,13 +351,20 @@ <h1>Evaluation<a class="headerlink" href="#evaluation" title="Link to this headi
 <td><p>9.0 GB</p></td>
 <td><p>37m13s</p></td>
 </tr>
-<tr class="row-odd"><td><p>gsplat-30k</p></td>
+<tr class="row-odd"><td><p>gsplat-30k (1 GPU)</p></td>
 <td><p>28.95</p></td>
 <td><p>0.870</p></td>
 <td><p>0.135</p></td>
 <td><p><strong>5.7 GB</strong></p></td>
 <td><p><strong>35m49s</strong></p></td>
 </tr>
+<tr class="row-even"><td><p>gsplat-30k (4 GPUs)</p></td>
+<td><p>28.91</p></td>
+<td><p>0.871</p></td>
+<td><p>0.135</p></td>
+<td><p><strong>2.0 GB</strong></p></td>
+<td><p><strong>11m28s</strong></p></td>
+</tr>
 </tbody>
 </table>
 </div>
@@ -779,7 +786,7 @@ <h2>Reproduced Metrics<a class="headerlink" href="#reproduced-metrics" title="Li
 is different from what’s reported in the original paper that uses
 <code class="code docutils literal notranslate"><span class="pre">from</span> <span class="pre">lpipsPyTorch</span> <span class="pre">import</span> <span class="pre">lpips</span></code>.</p>
 <p>The evaluation of <cite>gsplat-X</cite> can be reproduced with the command
-<code class="code docutils literal notranslate"><span class="pre">cd</span> <span class="pre">examples;</span> <span class="pre">bash</span> <span class="pre">benchmark.sh</span></code>
+<code class="code docutils literal notranslate"><span class="pre">cd</span> <span class="pre">examples;</span> <span class="pre">bash</span> <span class="pre">benchmarks/basic.sh</span></code>
 within the gsplat repo (commit 6acdce4).</p>
 <p>The evaluation of <cite>inria-X</cite> can be
 reproduced with our forked wersion of the official implementation at