AI Paper Atlas

Research map / 3D vision / 3D generation

DG-Recon: Depth-Guided Neural 3D Scene Reconstruction

Jihong Ju, Ching Wei Tseng, Oleksandr Bailo, Georgi Dikov, Mohsen Ghafoorian

ICCV 2023 · 3D vision › 3D generation · approach: Learning 3D reconstruction from images · 16 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add depth-guided back projection and occupancy mapping with average fusion (ablation)NeuralRecon baseline · ScanNet val subsetF-score: 0.590 -> 0.608Improved
add depth guidance with cross-attention fusion (ablation)cross-attention fusion · ScanNet val subsetF-score: 0.533 -> 0.608Improved
replace average fusion with variance fusion (ablation)DG-Recon · ScanNet val subsetF-score: 0.597 -> 0.629Improved
replace variance with cross-attention fusion and learnable initial query (ablation)DG-Recon · ScanNet val subsetF-score: 0.629 -> 0.640Improved
add depth offset d to cross-attention input features (ablation)c-att fusion · ScanNet val subsetF-score: 0.602 -> 0.634Improved
add ray direction r to cross-attention input features (ablation)c-att fusion · ScanNet val subsetF-score: 0.634 -> 0.640Improved
add auxiliary geometry features d and r to average fusion (ablation)average fusion · ScanNet val subsetF-score: 0.608 -> 0.597No clear effect
drop image features and use only geometry features d and r (ablation)c-att fusion · ScanNet val subsetF-score: 0.640 -> 0.595Hurt
replace learned reconstruction with non-learnable TSDF fusion of monocular depth priors (ablation)DG-Recon · ScanNet testF-score: 0.694 -> 0.580Hurt
use better monocular depth prior (Omnidata vs DPT with sparse inputs) (ablation)DG-Recon · 7-ScenesF-score: 0.542 -> 0.560Improved
use low-quality depth prior (Monodepth2) (ablation)DG-Recon · 7-ScenesF-score: 0.542 -> 0.252Hurt
use depth-guided DG-Recon (c-att) instead of NeuralReconNeuralRecon · ScanNet testFPS: 46 -> 20Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy