Research map / 3D vision / Depth and stereo
Look Deeper into Depth: Monocular Depth Estimation with Semantic Booster and Attention-Driven Loss
Jianbo Jiao, Ying Cao, Yibing Song, Rynson Lau
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 semi-dense up-skip connections (ablation) | depth-only network · NYUD2 (4-category semantics, architecture analysis) | rel: 0.157 -> 0.147 | Improved |
| add depth-aware loss L_DA (ablation) | depth network with SUC · NYUD2 (4-category semantics, architecture analysis) | rel: 0.147 -> 0.126 | Improved |
| add semantic labeling branch (no sharing) (ablation) | depth network with SUC and L_DA · NYUD2 (4-category semantics, architecture analysis) | rel: 0.126 -> 0.112 | Improved |
| add cross-stitch sharing between tasks (ablation) | depth+semantic network · NYUD2 (4-category semantics, architecture analysis) | rel: 0.112 -> 0.110 | Improved |
| use lateral sharing unit instead of cross-stitch (ablation) | depth+semantic network · NYUD2 (4-category semantics, architecture analysis) | rel: 0.110 -> 0.105 | Improved |
| add joint gradient loss and semantic focal loss (full attention-driven loss) on top of LSU (ablation) | depth+semantic network · NYUD2 (4-category semantics, architecture analysis) | rel: 0.105 -> 0.100 | Improved |
| use 40 semantic categories instead of 4 | Proposed network · NYUD2 test | rel: 0.100 -> 0.098 | Improved |
| add a semantic branch to Laina et al. | Laina et al. · NYUD2 test | rel: 0.127 -> 0.122 | Improved |
| add depth branch to semantic labeling (ablation) | Proposed network · NYUD2-40 semantic labeling | IoU: 48.9 -> 50.9 | Improved |
| add semantic focal loss to semantic-only baseline (ablation) | semantic labeling baseline · NYUD2-40 semantic labeling | IoU: 39.9 -> 48.9 | Improved |
| train with attention-driven loss and evaluate on truncated near depth ranges (ablation) | Proposed network · NYUD2 depth ranges | rel: 0.105 -> 0.100 | No clear effect |
Closest papers by meaning
- SDC-Depth: Semantic Divide-and-Conquer Network for Monocular Depth Estimation (CVPR 2020 · similarity 0.91)
- Guiding Monocular Depth Estimation Using Depth-Attention Volume (ECCV 2020 · similarity 0.91)
- Improving Monocular Depth Estimation by Leveraging Structural Awareness and Complementary Datasets (ECCV 2020 · similarity 0.89)
- Visualization of Convolutional Neural Networks for Monocular Depth Estimation (ICCV 2019 · similarity 0.89)
- Diffusion Models for Monocular Depth Estimation: Overcoming Challenging Conditions (ECCV 2024 · similarity 0.89)
- Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data (CVPR 2024 · 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.
