Research map / 3D vision / Depth and stereo
Multi-Loss Rebalancing Algorithm for Monocular Depth Estimation
Jae-Han Lee, Chang-Su Kim
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 weight rebalancing (lambda 3 to -3) with initialization to all 78 losses (ablation) | loss rebalancing depth estimator · NYUv2 | delta1: 86.4% -> 87.0% | Improved |
| add weight rebalancing (lambda 3 to -3) with initialization to all 78 losses (ablation) | loss rebalancing depth estimator · NYUv2 | RMSElin: 0.440 -> 0.434 | Improved |
| add weight initialization before rebalancing (ablation) | loss rebalancing depth estimator · NYUv2 | delta1: 85.6% -> 87.0% | Improved |
| add weight initialization alone, without rebalancing (ablation) | loss rebalancing depth estimator · NYUv2 | delta1: 86.4% -> 86.5% | No clear effect |
| use rebalancing on finest-scale losses only (L0) (ablation) | loss rebalancing depth estimator · NYUv2 | delta1: 85.3% -> 84.8% | Hurt |
| add rebalancing to coarse-scale losses L3 U L4 U L5 (ablation) | loss rebalancing depth estimator · NYUv2 | delta1: 86.6% -> 86.7% | Improved |
| use decreasing lambda 3 to -3 instead of constant lambda -3 (ablation) | loss rebalancing depth estimator · NYUv2 | delta1: 84.7% -> 87.0% | Improved |
| use decreasing lambda 3 to -3 instead of constant lambda 3 (ablation) | loss rebalancing depth estimator · NYUv2 | delta1: 86.6% -> 87.0% | Improved |
| widen lambda schedule to 5 to -5 instead of 3 to -3 (ablation) | loss rebalancing depth estimator · NYUv2 | delta1: 87.0% -> 86.4% | Hurt |
| replace uncertainty weighting [21] with proposed rebalancing (ablation) | loss rebalancing depth estimator · NYUv2 | delta1: 85.6% -> 87.0% | Improved |
| use all 78 losses instead of the single depth loss (ablation) | loss rebalancing depth estimator · NYUv2 | delta1: 84.8% -> 86.4% | Improved |
| add gradient losses to depth losses (multi-scale) (ablation) | loss rebalancing depth estimator · NYUv2 | delta1: 84.8% -> 86.5% | Improved |
Closest papers by meaning
- DME: Unveiling the Bias for Better Generalized Monocular Depth Estimation (AAAI 2024 · similarity 0.87)
- Rethinking Depth Estimation for Multi-View Stereo: A Unified Representation (CVPR 2022 · similarity 0.87)
- Single View Stereo Matching (CVPR 2018 · similarity 0.87)
- Revisiting Gradient-Based Uncertainty for Monocular Depth Estimation (TPAMI 2025 · similarity 0.87)
- Unsupervised Monocular Depth Estimation With Left-Right Consistency (CVPR 2017 · similarity 0.87)
- Monocular Depth Estimation Using Whole Strip Masking and Reliability-Based Refinement (ECCV 2018 · 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.
