SeProRS - UCM - Train 80%: leaderboard
SeProRS - UCM - Train 80%. SeProRS arXiv2611.13851v1 TablesV-VII: remote-sensing scene classifiers trained separately for each dataset/split. Text pipeline frozen bge-base with trained classifier on one-time VLM semantic proxies; G1=LLaVA-OneVision-Qwen2-0.5B,G2=Qwen2.5-VL-3B,G3=Gemini2.5Flash. Flat/Structured prompts are separate pipelines. Published means and standard deviations across seeds; source-specific trained pipeline identities, not scores assigned to the frozen generator alone. Vision classifiers are dataset-trained method runs, not newly released general-purpose checkpoints. Metric: Test classification accuracy (%). Higher is better.
Metric: Test classification accuracy (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | MSCN (Benchmark Sweep Run: source=SeProRS; dataset=UCM; training=80%; pipeline=dataset-trained-classifier) | 99.95 | |
| 2 | CGINet (Benchmark Sweep Run: source=SeProRS; dataset=UCM; training=80%; pipeline=dataset-trained-classifier) | 99.84 | |
| 3 | DINOv3-Adapter (Benchmark Sweep Run: source=SeProRS; dataset=UCM; training=80%; pipeline=dataset-trained-classifier) | 99.52 | |
| 4 | HFAM (Benchmark Sweep Run: source=SeProRS; dataset=UCM; training=80%; pipeline=dataset-trained-classifier) | 99.43 | |
| 5 | UniRepLKNet (Benchmark Sweep Run: source=SeProRS; dataset=UCM; training=80%; pipeline=dataset-trained-classifier) | 99.24 | |
| 6 | UPetu (Benchmark Sweep Run: source=SeProRS; dataset=UCM; training=80%; pipeline=dataset-trained-classifier) | 99.05 | |
| 7 | G3 + Flat (Benchmark Sweep Run: source=SeProRS; dataset=UCM; training=80%; pipeline=dataset-trained-classifier) | 95.39 | |
| 8 | G3 + Structured (Benchmark Sweep Run: source=SeProRS; dataset=UCM; training=80%; pipeline=dataset-trained-classifier) | 92.16 | |
| 9 | G1 + Flat (Benchmark Sweep Run: source=SeProRS; dataset=UCM; training=80%; pipeline=dataset-trained-classifier) | 91.24 | |
| 10 | G1 + Structured (Benchmark Sweep Run: source=SeProRS; dataset=UCM; training=80%; pipeline=dataset-trained-classifier) | 89.48 |
Interactive version: theaggregate.ai/benchmark?slug=seprors-ucm-train-80 · How It Works · Data refreshed daily, snapshot 2026-11-13.