Research map / Image generation / Image restoration
Self-Supervised Underwater Caustics Removal and Descattering via Deep Monocular SLAM
Jonathan Sauder, Devis Tuia
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 |
|---|---|---|---|
| apply CausticsNet before COLMAP feature matching | COLMAP image matching on underwater reef videos · COLMAP matched keypoints (6 scenes, total) | geometrically verified matched keypoints: 4512 -> 6085 | Improved |
| apply BackscatterNet alone before COLMAP matching | COLMAP image matching on underwater reef videos · COLMAP matched keypoints (6 scenes, total) | geometrically verified matched keypoints: 4512 -> 4572 | No clear effect |
| add BackscatterNet on top of CausticsNet (ablation) | CausticsNet preprocessing for COLMAP matching · COLMAP matched keypoints (6 scenes, total) | geometrically verified matched keypoints: 6085 -> 5696 | Hurt |
| use classical underwater enhancement methods (e.g. UDCP, GBdehazing) instead of raw RGB | COLMAP image matching · COLMAP matched keypoints (6 scenes, total) | average improvement over RGB: 0.71 (UDCP), 0.71 (GBdehazingRCorrection) vs 1.0 for RGB | Hurt |
| use per-split thresholds instead of one global threshold to binarize predicted caustics (ablation) | CausticsNet on R-Caustics · R-Caustics (all splits) | F1: 68.61 -> 72.37 | Improved |
| zero-shot self-supervised CausticsNet vs supervised DeepLabV3+ trained on R-Caustics | caustics segmentation · R-Caustics (test splits 1,2,3) | F1: 77.71 vs 78.62 | No clear effect |
| zero-shot CausticsNet vs supervised baseline on splits 5,7 | caustics segmentation · R-Caustics (test splits 5,7) | F1: 67.20 vs 79.72 | Hurt |
Closest papers by meaning
- Sea-Thru: A Method for Removing Water From Underwater Images (CVPR 2019 · similarity 0.86)
- DRM-Net: Explicit Residual Modelling with Subaquatic Multi-Scale Context Fusion for Underwater Image Enhancement (AAAI 2026 · similarity 0.86)
- WaterDiffusion: Learning a Prior-involved Unrolling Diffusion for Joint Underwater Saliency Detection and Visual Restoration (AAAI 2025 · similarity 0.85)
- Atlantis: Enabling Underwater Depth Estimation with Stable Diffusion (CVPR 2024 · similarity 0.84)
- Sea-ing in Low-light (CVPR 2025 · similarity 0.84)
- NAUTILUS: A Large Multimodal Model for Underwater Scene Understanding (NeurIPS 2025 · similarity 0.83)
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.
