Shallow-Water UAV Assessment - K - Presence Balanced Accuracy: leaderboard
Shallow-water UAV distortion assessment v1 TableVI. KAUSTics K=600 and Sea-Undistort S=400 images. Each model uses its best pooled presence-balanced-accuracy prompt (P0-P3); that same prompt is used across both datasets and metrics. Presence BA is balanced accuracy; joint 1-RPS is the native full-denominator ordinal skill score. Source confidence intervals remain in the pinned table. Conditional answered-only metrics, valid-output coverage, resource costs and every restoration/image-generation metric are withheld.
Metric: Native bounded score (0–1). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | SenseNova-U1 [48] (P1) (Shallow-Water UAV Run: prompt=best-pooled-presence-BA;dataset=K (600);effort=unreported) | 0.58 | |
| 2 | MammothModa2 [46] (P1) (Shallow-Water UAV Run: prompt=best-pooled-presence-BA;dataset=K (600);effort=unreported) | 0.55 | |
| 3 | BAGEL [44] (P0) (Shallow-Water UAV Run: prompt=best-pooled-presence-BA;dataset=K (600);effort=unreported) | 0.54 | |
| 4 | InternVL-U [45] (P3) (Shallow-Water UAV Run: prompt=best-pooled-presence-BA;dataset=K (600);effort=unreported) | 0.54 | |
| 5 | S1-Omni-Image [47] (P1) (Shallow-Water UAV Run: prompt=best-pooled-presence-BA;dataset=K (600);effort=unreported) | 0.5 |
Interactive version: theaggregate.ai/benchmark?slug=shallow-water-uav-assessment-k-presence-balanced-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-12.