Research map / 3D vision / Depth and stereo
RobustSpring: Benchmarking Robustness to Image Corruptions for Optical Flow, Scene Flow and Stereo
Victor Oei, Jenny Schmalfuss, Lukas Mehl, Madlen Bartsch, Shashank Agnihotri, Margret Keuper, Andreas Bulling, Andres Bruhn
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 |
|---|---|---|---|
| exclude rain pixels (about 90% of image) from the robustness score (ablation) | RobustSpring metric · RobustSpring optical flow | R^c_EPE rain: score changes by at most 5% | No clear effect |
| use RobustSpring subsampling (0.05%) instead of Spring's original 1% subsampling (ablation) | RobustSpring metric · RobustSpring optical flow | R^c_EPE: SEA-RAFT 3.19 -> 2.96 (matches full data 2.96) | Improved |
| subsample Hero-frames in addition (ablation) | RobustSpring metric · RobustSpring optical flow | R^c_EPE: Spring 3.19 vs Spring* 2.96 for SEA-RAFT | Improved |
| use average vs median vs Schulze ranking of models (ablation) | RobustSpring metric · RobustSpring optical flow | model ranking: GMFlow rank 2 (Average) vs 5 (Median, Schulze) | Mixed |
| more accurate optical flow models | optical flow models · RobustSpring | R^c_EPE: trend weak and nonlinear | Mixed |
| more accurate optical flow models under noise corruptions | optical flow models · RobustSpring noise | R^c_EPE noise | Hurt |
| more accurate optical flow models under weather corruptions | optical flow models · RobustSpring weather | R^c_EPE weather | Improved |
| use transformer architecture (GMFlow, FlowFormer) | optical flow · RobustSpring noise | R^c_EPE noise | Hurt |
| use stacked/progressive refinement (SEA-RAFT, FlowNet2) | optical flow · RobustSpring noise | R^c_EPE noise: SEA-RAFT Gaussian noise 1.49; FlowNet2 1.33 | Improved |
| use scene flow models instead of conventional optical flow | scene flow · RobustSpring | R^c_EPE | No clear effect |
| evaluate on noisier real KITTI frames instead of RobustSpring noise | optical flow · KITTI noisiest 10% | relative noise robustness: rankings transfer except FlowFormer | Improved |
| use M-FUSE rather than RAFT-3D for optical flow | scene flow · RobustSpring | Average R^c_EPE: 5.03 -> 3.39 | Improved |
Closest papers by meaning
- VLM-RobustBench: A Comprehensive Benchmark for Robustness of Vision-Language Models (ICML 2026 · similarity 0.88)
- LAION-C: An Out-of-Distribution Benchmark for Web-Scale Vision Models (ICML 2025 · similarity 0.88)
- Shape and Texture: What Influences Reliable Optical Flow Estimation? (CVPR 2025 · similarity 0.87)
- REOBench: Benchmarking Robustness of Earth Observation Foundation Models (NeurIPS 2025 · similarity 0.87)
- Scalable Benchmarking and Robust Learning for Noise-Free Ego-Motion and 3D Reconstruction from Noisy Video (ICLR 2025 · similarity 0.87)
- ImageNet-D: Benchmarking Neural Network Robustness on Diffusion Synthetic Object (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.
