Research map / Image generation / Text-to-image diffusion
WaterGen: Decoupling Scene and Medium in Underwater Image Generation
Jiayi Wu, Tianfu Wang, Tianyi Xiong, Dehao Yuan, Xiaomin Lin, Md Jahidul Islam, Cornelia Fermuller, Christopher Metzler, Yiannis Aloimonos
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 |
|---|---|---|---|
| Use decoupled WaterGen scene generation instead of Atlantis | WaterGen · Clear scene generation | UIQM: 2.8338 -> 3.0239 | Improved |
| Use WaterGen instead of TIDE | WaterGen · Clear scene generation | MUSIQ: 66.4304 -> 69.2638 | Improved |
| Use WaterGen instead of Atlantis for text alignment | WaterGen · Clear scene generation | CLIP score: 0.2457 -> 0.2614 | Improved |
| Add WaterGen synthetic data to restoration training | WaterNet, Phaseformer, DeepWaveNet, Histoformer · UIIS10K | MUSIQ: 66.866 -> 68.322 (WaterNet) | Improved |
| Add WaterGen synthetic data to restoration training | Restoration models · UIIS10K | UIQM: 2.239 -> 2.272 (Phaseformer) | Improved |
| Add WaterGen data to real data for segmentation | SegFormer (MiT-B4) · UIIS | mIoU: 70.2 -> 75.6 | Improved |
| Add WaterGen data to real data for segmentation | Mask2Former (Swin-B) · USIS10K | mIoU: 76.1 -> 76.9 | Improved |
| Use WaterGen data instead of SynTIDE as extra segmentation data | Segmentation models · UIIS / USIS10K | mIoU: Mask2Former UIIS 74.3 vs 74.0 | Mixed |
| Inject medium via decoder instead of ControlNet (ablation) | WaterGen · Medium control accuracy | CIEDE2000: 43.70 -> 5.44 | Improved |
| Inject medium via decoder instead of T2I-Adapter (ablation) | WaterGen · Medium control accuracy | Mean angular error: 9.96 -> 4.42 | Improved |
| Add stochastic medium noise injection in decoder training (ablation) | WaterGen decoder · Medium control accuracy | CIEDE2000: 7.27 -> 5.44 | Improved |
| Add bidirectional UIFM consistency loss (ablation) | WaterGen decoder · Medium control accuracy | Mean angular error: 7.31 -> 4.42 | Improved |
Closest papers by meaning
- Atlantis: Enabling Underwater Depth Estimation with Stable Diffusion (CVPR 2024 · similarity 0.89)
- WaterDiffusion: Learning a Prior-involved Unrolling Diffusion for Joint Underwater Saliency Detection and Visual Restoration (AAAI 2025 · similarity 0.89)
- Diving into Underwater: Segment Anything Model Guided Underwater Salient Instance Segmentation and A Large-scale Dataset (ICML 2024 · similarity 0.88)
- Synergistic Multiscale Detail Refinement via Intrinsic Supervision for Underwater Image Enhancement (AAAI 2024 · similarity 0.88)
- A Unified Image-Dense Annotation Generation Model for Underwater Scenes (CVPR 2025 · similarity 0.88)
- NAUTILUS: A Large Multimodal Model for Underwater Scene Understanding (NeurIPS 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.
