You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
feat: add quantitative metrics tables with all baseline comparisons
- Add DatasetMetrics and SceneMetrics types
- Include AsyncEv-Deblur (7 scenes) and Ev-DeblurBlender (4 scenes) datasets
- Compare 5 methods: 3DGS, BAGS, DeblurGS, LSENeRF, Ours
- Display PSNR, SSIM, LPIPS metrics with color-coded best values
- Blue for best PSNR, Yellow for best SSIM, Green for best LPIPS
- Include Average row for each dataset
<h3className="text-lg font-medium mb-4 text-center">PSNR over Training Epochs</h3>
588
-
<MetricsChartdata={metrics}/>
589
-
</div>
590
-
<divclassName="flex-1 text-slate-600 space-y-4">
591
-
<p>
592
-
Our model demonstrates rapid convergence, achieving a <strong>PSNR of {metrics[metrics.length-1].psnr}dB</strong> by epoch 50.
593
-
The combination of our novel loss function and the initialized Gaussian field significantly accelerates the training process compared to baseline NeRF methods.
<li>Consistent performance across diverse datasets.</li>
599
-
</ul>
600
-
</div>
601
-
</div>
605
+
<pclassName="text-slate-600 mb-6">
606
+
We compare our method against state-of-the-art approaches including Original 3D Gaussian Splatting, BAGS, DeblurringGS, and LSE-NeRF across two datasets.
607
+
Our method achieves the best performance on most scenes, with significant improvements in PSNR, SSIM, and LPIPS metrics.
0 commit comments