Research map / 3D vision / Depth and stereo
iDisc: Internal Discretization for Monocular Depth Estimation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 internal scene discretization (ISD) in place of explicit depth discretization (EDD) (ablation) | iDisc without MSDA/AFP · NYU official test | RMS: 0.367 -> 0.340 | Helped |
| add adaptive feature partitioning (AFP) on top of ISD (ablation) | iDisc · NYU official test | RMS: 0.340 -> 0.319 | Helped |
| add multi-scale deformable attention (MSDA) on top of explicit discretization (ablation) | EDD model · NYU official test | RMS: 0.367 -> 0.326 | Helped |
| add MSDA on top of ISD + AFP (ablation) | iDisc · NYU official test | RMS: 0.319 -> 0.313 | Helped |
| random initialization of IDRs and architecture similar to prior work instead of learnable priors (ablation) | AFP · NYU official test | RMS: 0.313 -> 0.319 | Hurt |
| use fewer IDRs in the sweep (2-128); optimum in [8,32] (ablation) | iDisc · NYU ablation over number of IDRs | qualitative trend | No clear effect |
| increase AFP iterations from 2 to 3 or 4, or ISD layers from 2 to 3 or 4 (ablation) | iDisc · NYU official test | RMS: 0.313-0.317 | No clear effect |
| removing MSDA from iDisc | iDisc · Computational throughput on RTX 3090 | throughput (fps): 8.3 -> 20.7 fps | Helped |
| iDisc with SwinL vs NeWCRF SwinL on NYU | depth estimation · NYU official test | RMS: 0.334 -> 0.313 | Helped |
| zero-shot from KITTI to Argoverse and DDAD | iDisc trained on KITTI · Zero-shot outdoor | δ1: Argoverse 0.560 vs BTS 0.307; DDAD 0.350 vs BTS 0.399 | Mixed |
Closest papers by meaning
- A Simple yet Universal Framework for Depth Completion (NeurIPS 2024 · similarity 0.90)
- Depth2Pose: A Pose-Based Benchmark for Monocular Depth Estimation without Ground-Truth Depth (NeurIPS 2026 · similarity 0.90)
- DME: Unveiling the Bias for Better Generalized Monocular Depth Estimation (AAAI 2024 · similarity 0.89)
- Depth Pro: Sharp Monocular Metric Depth in Less Than a Second (ICLR 2025 · similarity 0.88)
- Mind The Edge: Refining Depth Edges in Sparsely-Supervised Monocular Depth Estimation (CVPR 2024 · similarity 0.88)
- GeoDepth: From Point-to-Depth to Plane-to-Depth Modeling for Self-Supervised Monocular Depth Estimation (CVPR 2025 · similarity 0.88)
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.