Research map / Architectures and efficiency / Neural architecture search
AutoDispNet: Improving Disparity Estimation With AutoML
Tonmoy Saikia, Yassine Marrakchi, Arber Zela, Frank Hutter, Thomas Brox
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 |
|---|---|---|---|
| replace DispNet-C with architecture found by Dense-DARTS | DispNet-C · Sintel clean (train) | EPE: 3.19 -> 2.85 | Improved |
| tune learning rate and weight decay with BOHB (ablation) | AutoDispNet-C · Sintel clean (train) | EPE: 2.85 -> 2.66 | Improved |
| tune learning rate and weight decay with BOHB (ablation) | AutoDispNet-C · FlyingThings3D test | EPE: 1.53 -> 1.51 | No clear effect |
| stack a refinement network (CS) with Dense-DARTS | AutoDispNet-C · Sintel clean (train) | EPE: 2.85 -> 2.30 | Improved |
| add a second refinement network (CSS) | AutoDispNet-CS · Sintel clean (train) | EPE: 2.30 -> 2.14 | Improved |
| tune hyperparameters with BOHB (ablation) | AutoDispNet-CSS · Sintel clean (train) | EPE: 2.14 -> 1.97 | Improved |
| use AutoDispNet-cs (2 nets) instead of DispNet-CSS (3 nets) | DispNet-CSS · Sintel clean (train) | EPE: 2.36 -> 2.30 | Improved |
| shrink C_init to 18 (css) (ablation) | AutoDispNet-CSS · Sintel clean (train) | EPE: 2.14 -> 2.54 | Hurt |
| learn new cells for refinement network with Dense-DARTS (ablation) | AutoDispNet-CS · Sintel clean (train) | EPE: 2.30 -> 2.32 | No clear effect |
| tune new hyperparameters for refinement network with BOHB (ablation) | AutoDispNet-CS with BOHB · Sintel clean (train) | EPE: 2.14 -> 2.16 | No clear effect |
| replace Laina et al. with Dense-DARTS network | Laina et al. · SUN3D single-view depth | Abs. rel: 0.272 -> 0.234 | Improved |
| replace Laina et al. with BOHB-tuned Dense-DARTS network | Laina et al. · NYU-Depth-V2 | Abs. rel: 0.127 -> 0.170 | Hurt |
Closest papers by meaning
- BRNet: Exploring Comprehensive Features for Monocular Depth Estimation (ECCV 2022 · similarity 0.86)
- AutoFormer: Searching Transformers for Visual Recognition (ICCV 2021 · similarity 0.86)
- Self-Supervised Monocular Depth Estimation by Direction-aware Cumulative Convolution Network (ICCV 2023 · similarity 0.86)
- LocalBins: Improving Depth Estimation by Learning Local Distributions (ECCV 2022 · similarity 0.85)
- Self-Supervised Monocular Trained Depth Estimation Using Self-Attention and Discrete Disparity Volume (CVPR 2020 · similarity 0.85)
- Closest Neighbors are Harmful for Lightweight Masked Auto-encoders (CVPR 2025 · similarity 0.84)
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.
