Skip to content

Commit

Permalink
comparison text 4/5
Browse files Browse the repository at this point in the history
  • Loading branch information
@w-m
w-m committed Jul 9, 2024
1 parent ebded6f commit 83f7ef3
Showing 1 changed file with 48 additions and 37 deletions.
85 changes: 48 additions & 37 deletions project-page/index.html
View file
Original file line number Diff line number Diff line change
Expand Up @@ -321,18 +321,35 @@ <h2 class="title is-5">Flowers (Mip-NeRF 360 dataset)</h2>
</tr>
</table>
</div>
</div>
</div>
</section>

<h2 class="title is-3">Comparison with other 3D scene representations</h2>
<div class="content has-text-justified">
<section class="section hero is-small">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Comparison with other 3D scene representations</h2>
<div class="content has-text-justified">
<strong>Quantitative results:</strong> Our method significantly outperforms the default 3D Gaussian Splatting (3DGS) and prior NeRF-based 3D
reconstruction methods in terms of storage size and rendering efficiency. We achieve a <b>reduction in storage size</b> by a factor of
<b>17x to 42x</b>, depending on the dataset, with the most notable reduction being 41.6x on the Deep Blending dataset. Deactivating spherical
harmonics during training further enhances this reduction to 127x over vanilla 3DGS, while also improving PSNR metrics. Our approach maintains
competitive visual quality metrics (PSNR and L-PIPS) compared to Mip-NeRF360 and VQ-TensoRF but allows for real-time rendering and significantly
faster training, approximately 10 to 30 minutes per scene. Additionally, our optimized method renders scenes more quickly, achieving 515 fps
compared to 385 fps for the vanilla 'Truck' dataset model on the same GPU, while also displaying better visual quality, compared to vanilla 3DGS.
<div class="content has-text-justified">
<strong>Quantitative results:</strong> Our method significantly outperforms the default 3D Gaussian Splatting (3DGS) and prior NeRF-based 3D
reconstruction methods in terms of storage size and rendering efficiency. We achieve a <b>reduction in storage size</b> by a factor of
<b>17x to 42x</b>, depending on the dataset, with the most notable reduction being 41.6x on the Deep Blending dataset. Deactivating spherical
harmonics during training further enhances this reduction to 127x over vanilla 3DGS, while also improving PSNR metrics. Our approach maintains
competitive visual quality metrics (PSNR and L-PIPS) compared to Mip-NeRF360 and VQ-TensoRF but allows for real-time rendering and significantly
faster training, approximately 10 to 30 minutes per scene. Additionally, our optimized method renders scenes more quickly, achieving 515 fps
compared to 385 fps for the vanilla 'Truck' dataset model on the same GPU, while also displaying better visual quality, compared to vanilla
3DGS.
</div>
</div>
</div>
</div>
</div>
</section>
<section class="hero is-small">
<div class="hero-body">
<div class="container">
<div class="content has-text-justified">
<img style="display: block; margin-left: auto; margin-right: auto; width: 90%" src="static/images/table.png" alt="MY ALT TEXT" />
</div>
</div>
Expand Down Expand Up @@ -365,34 +382,28 @@ <h2 class="title is-3">Comparison with other 3DGS compression methods</h2>
<p>We invite you to explore other methods for compressing Gaussian splats and compare their approaches in this survey:</p>
</div>

<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<div class="columns">
<div class="column is-one-third">
<img src="static/images/survey-teaser.png" alt="" />
</div>
<div class="column is-two-thirds has-text-justified">
<span class="link-block">
<a href="https://survey.3dgs.zip/" target="_blank" class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fas fa-chart-line"></i>
</span>
<span>Survey Page</span>
</a>
<a href="https://github.com/w-m/3dgs-compression-survey" target="_blank" class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fab fa-github"></i>
</span>
<span>Survey Code & Data</span>
</a>
</span>
<div>
<b>3dgs.zip: A survey on 3D Gaussian Splatting Compression Methods</b>: We're collecting results and info on methods that reduce the
storage size of 3D Gaussian splatting scenes. Contributions welcome!
</div>
</div>
</div>
<div class="columns">
<div class="column is-one-third">
<img src="static/images/survey-teaser.png" alt="" />
</div>
<div class="column is-two-thirds has-text-justified">
<span class="link-block">
<a href="https://survey.3dgs.zip/" target="_blank" class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fas fa-chart-line"></i>
</span>
<span>Survey Page</span>
</a>
<a href="https://github.com/w-m/3dgs-compression-survey" target="_blank" class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fab fa-github"></i>
</span>
<span>Survey Code & Data</span>
</a>
</span>
<div>
<b>3dgs.zip: A survey on 3D Gaussian Splatting Compression Methods</b>: We're collecting results and info on methods that reduce the storage
size of 3D Gaussian splatting scenes. Contributions welcome!
</div>
</div>
</div>
Expand Down

0 comments on commit 83f7ef3

Please sign in to comment.