Research map / 3D vision / Depth and stereo
PacGDC: Label-Efficient Generalizable Depth Completion with Projection Ambiguity and Consistency
Haotian Wang, Aoran Xiao, Xiaoqin Zhang, Meng Yang, Shijian Lu
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 |
|---|---|---|---|
| add DepthAnything pseudo depth labels (ablation) | SPNet-Tiny · Ablation 25% data | RMSE (mm): 2484 -> 2463 | Improved |
| raise interpolation probability from 0.0 to 1.0 (ablation) | SPNet-Tiny · Ablation 25% data | RMSE (mm): 2463 -> 2330 | Improved |
| add relocation of interpolated labels (ablation) | SPNet-Tiny · Ablation 25% data | RMSE (mm): 2330 -> 2277 | Improved |
| add DepthPro as second foundation model (ablation) | SPNet-Tiny · Ablation 25% data | RMSE (mm): 2277 -> 2241 | Improved |
| add 390K unlabeled SA1B images (ablation) | SPNet-Tiny · Ablation 25% data | RMSE (mm): 2241 -> 2143 | Improved |
| apply full PacGDC synthesis to SPNet-Tiny (ablation) | SPNet-Tiny · Ablation 25% data | MAE (mm): 990 -> 792 | Improved |
| apply full PacGDC synthesis to G2-MonoDepth (ablation) | G2-MonoDepth · Ablation 25% data | RMSE (mm): 2669 -> 2192 | Improved |
| add DistillAnyDepth to DA+DepthPro+DAV2 foundation model set (ablation) | SPNet-Tiny · Ablation 25% data | RMSE (mm): 2232 -> 2279 | Hurt |
| add DAV2 to DA+DepthPro foundation model set (ablation) | SPNet-Tiny · Ablation 25% data | RMSE (mm): 2241 -> 2232 | Improved |
| PacGDC synthesis vs SPNet baseline on uniform sparse depth zero-shot (Large) | SPNet-Large · Zero-shot uniform mean | RMSE (mm): 2271 -> 1966 | Improved |
| PacGDC pretrained weights for few-shot fine-tuning vs best baseline (1 sample) | PacGDC · KITTI 64-line few-shot, 1 sample | RMSE (mm): 1358 -> 1078 | Improved |
| 3x3 instead of 2x2 convolution in SPNet up/down sampling | SPNet · inference cost | speed (image/s): 126.6 -> 121.8 | Hurt |
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.
- G2-MonoDepth — G2-MonoDepth: A General Framework of Generalized Depth Inference From Monocular RGB+X Data (TPAMI 2023) · better on 4 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
- A Simple yet Universal Framework for Depth Completion (NeurIPS 2024 · similarity 0.90)
- Zero-shot Depth Completion via Test-time Alignment with Affine-invariant Depth Prior (AAAI 2025 · similarity 0.89)
- DepthFocus: Controllable Depth Estimation for See-Through Scenes (CVPR 2026 · similarity 0.88)
- Zero-Shot Novel View and Depth Synthesis with Multi-View Geometric Diffusion (CVPR 2025 · similarity 0.87)
- Depth2Pose: A Pose-Based Benchmark for Monocular Depth Estimation without Ground-Truth Depth (NeurIPS 2026 · similarity 0.87)
- SharpDepth: Sharpening Metric Depth Predictions Using Diffusion Distillation (CVPR 2025 · similarity 0.87)
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.
