Research map / 3D vision / Depth and stereo
Virtual Occlusions Through Implicit Depth
Jamie Watson, Mohamed Sayed, Zawar Qureshi, Gabriel J. Brostow, Sara Vicente, Oisin Mac Aodha, Michael Firman
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 changed | Where | Measured | Effect |
|---|---|---|---|
| reframe occlusion as direct binary mask prediction (implicit depth) instead of depth regression then threshold | SimpleRecon backbone · ScanNetv2 test keyframes | IoU All: +2.09 (60.52 to 62.61) | Improved |
| reframe occlusion as direct binary mask prediction instead of depth regression | SimpleRecon (ResNet) backbone · ScanNetv2 test keyframes | IoU All: +1.23 (58.91 to 60.14) | Improved |
| implicit depth occlusion on MonoDepth2 backbone | MonoDepth2 · ScanNetv2 test keyframes | IoU All: +2.29 (46.06 to 48.35) | Improved |
| implicit depth occlusion on ManyDepth backbone | ManyDepth · ScanNetv2 test keyframes | IoU All: +0.87 (55.68 to 56.55) | Improved |
| implicit depth occlusion on SimpleRecon backbone | SimpleRecon · ScanNetv2 test keyframes | IoU Surface: +3.04 (32.44 to 35.48) | Improved |
| implicit depth occlusion on SimpleRecon backbone | SimpleRecon · ScanNetv2 test keyframes | Abs Rel (depth, via binary search): 0.0436 vs 0.0429 (slightly worse) | Mixed |
| remove edge-based regularization near depth discontinuities (ablation) | SimpleRecon (ResNet) + Ours · ScanNetv2 test keyframes | IoU All: adding it: 59.79 to 60.14 | Improved |
| remove high-resolution supervision (ablation) | SimpleRecon (ResNet) + Ours · ScanNetv2 test keyframes | IoU All: adding it: 59.94 to 60.14 (small) | Improved |
| replace multi-view backbone with monocular backbone (ablation) | implicit depth model · ScanNetv2 test keyframes | IoU All: 48.24 vs 60.14 without MVS | Hurt |
| replace implicit binary output with DORN-style 80-bin classification (ablation) | SimpleRecon (ResNet) backbone · ScanNetv2 test keyframes | IoU All: 56.81 vs 60.14 | Hurt |
| uniform training depth sampling instead of near-surface biased sampling (ablation) | implicit depth model · ScanNetv2 test keyframes | IoU All: 60.28 vs 60.14 overall, but IoU Surface 33.18 vs 33.29 and Boundary 36.16 vs 36.54 | Mixed |
| add warped previous prediction temporal input (with training-time noise and flip) (ablation) | implicit depth model · ScanNetv2 test keyframes | Temporal Score (lower is better): 235.1 to 164.5 (about 30% fewer flickers) | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SimpleRecon — SimpleRecon: 3D Reconstruction without 3D Convolutions (ECCV 2022) · better on 3 of 4 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
- Depth Estimation with Occlusion Modeling Using Light-Field Cameras (TPAMI 2016 · similarity 0.86)
- A Novel Occlusion-aware Vote Cost for Light Field Depth Estimation (TPAMI 2021 · similarity 0.85)
- Efficient Depth Estimation for Unstable Stereo Camera Systems on AR Glasses (CVPR 2025 · similarity 0.84)
- Seurat: From Moving Points to Depth (CVPR 2025 · similarity 0.83)
- RadOcc: Learning Cross-Modality Occupancy Knowledge through Rendering Assisted Distillation (AAAI 2024 · similarity 0.83)
- ODG: Occupancy Prediction Using Dual Gaussians (NeurIPS 2025 · similarity 0.82)
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.
