Scribd
Upload
27 views1 page

Results

Uploaded by

Maaz Sayyed
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
27 views1 page

Results

Uploaded by

Maaz Sayyed
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 1

Method Abs Rel ↓ Abs Diff ↓ Sq Rel ↓ RMSE ↓ δ < 1.

25 ↑ Comp ↑
SimpleRe 0.212 0.109 0.448 0.207 0.952 0.983
c
NeuralRec 0.194 0.125 0.322 0.231 0.932 0.871
Ours 0.215 0.109 0.454 0.220 0.942 0.949

Table 1. Quantitative results of 2D metrics on 7-Scenes. We evaluate our method on the official test
split of 7-scenes using the same 2D metrics .All methods are trained on ScanNet and for baseline
methods, we use their released pre-trained models.

Method Acc ↓ Fusion


Depth Comp ↓ Chamfer ↓ Prec ↑ Recall ↑ F-score ↑
COLMAP 10.22Depth 11.88
Fusion 11.05 0.509 0.474 0.489
MVDNet 12.94 8.34 10.64 0.443 0.487 0.460
SimpleRe 5.53 6.09 5.81 0.686 0.658 0.671
c
Feature Fusion
3DVNet 7.72 6.73 7.22 0.655 0.596 0.621
Atlas 7.16 7.61 0.675 0.251 0.605 0.636
Feature Fusion
Depth FsionNeuralRec 5.09 9.13 7.11 0.630 0.612 0.619
Ours 4.17 9.05 6.61 0.751 0.580 0.653

Table 4. Quantitative results of 3D metrics on ScanNet following the evaluation protocol in [1]. We
show the results of two stage depth fusion methods (top) and those for end-to-end feature fusion
works (bottom). We highlight the best results for Depth Fusion, Feature Fusion method in blue, teal,
and violet, respectively.. Offline methods assume to observe the whole video sequence. Compared
to all existing online feature fusion methods [1, 14], our method achieves the best performance in
the Chamfer distance metric.

You might also like