Research map / Label-efficient and robust learning / Self-supervised and contrastive
ROIFormer: Semantic-Aware Region of Interest Transformer for Efficient Self-Supervised 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 |
|---|---|---|---|
| use local semantic-aware ROI attention instead of Transformer or deformable attention | ROIFormer depth estimation · KITTI, Res50, 1024 resolution | Sq Rel: 0.7383 -> 0.6161 | Helped |
| add ROI attention to monocular depth features (ablation) | self-supervised monocular depth estimation · KITTI component ablation | AbsRel: 0.115 -> 0.108 | Helped |
| add semantic information to ROIFormer (ablation) | self-supervised monocular depth estimation · KITTI component ablation | AbsRel: 0.108 -> 0.1005 | Helped |
| add semantic-guided reprojection mask loss (ablation) | ROIFormer with semantic guidance · KITTI component ablation | AbsRel: 0.1005 -> 0.1002 | Helped |
| add 1 ROI attention layer (ablation) | ROIFormer · KITTI depth ablation | AbsRel: 1.037 -> 1.011 | Helped |
| use 2 rather than 1 ROI attention layer (ablation) | ROIFormer · KITTI depth ablation | AbsRel: 1.011 -> 1.002 | Helped |
| increase ROI attention depth from 2 to 3 layers (ablation) | ROIFormer · KITTI depth ablation | AbsRel: 1.002 -> 1.015 | Hurt |
| increase ROI attention depth from 3 to 4 layers (ablation) | ROIFormer · KITTI depth ablation | AbsRel: 1.015 -> 1.027 | Hurt |
| add feature fusion at P4, P3 and P2 in succession (ablation) | ROIFormer feature pyramid · KITTI depth ablation | AbsRel: 0.103 -> 0.1008 -> 0.1003 -> 0.1002 | Helped |
| use [8,16,32] rather than [4,8,16] attention points across pyramid levels (ablation) | ROIFormer · KITTI attention-point ablation | AbsRel: 0.1010 -> 0.1003 | Helped |
| increase multi-level attention points from [8,16,32] to [16,32,64] (ablation) | ROIFormer · KITTI attention-point ablation | AbsRel: 0.1003 -> 0.0998 | Helped |
| use 32 rather than 16 fixed attention points (ablation) | ROIFormer · KITTI attention-point ablation | AbsRel: 0.1009 -> 0.1012 | Hurt |
Closest papers by meaning
- Semantically-Guided Representation Learning for Self-Supervised Monocular Depth (ICLR 2020 · similarity 0.91)
- Multi-Frame Self-Supervised Depth With Transformers (CVPR 2022 · similarity 0.90)
- Self-Supervised Monocular Trained Depth Estimation Using Self-Attention and Discrete Disparity Volume (CVPR 2020 · similarity 0.89)
- Improved Monocular Depth Prediction Using Distance Transform Over Pre-semantic Contours with Self-supervised Neural Networks (CVPR 2025 · similarity 0.88)
- Mining Supervision for Dynamic Regions in Self-Supervised Monocular Depth Estimation (CVPR 2024 · similarity 0.86)
- Know Your Neighbors: Improving Single-View Reconstruction via Spatial Vision-Language Reasoning (CVPR 2024 · similarity 0.86)
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.