Cloud Optical Thickness (COT) Regression on HyperFM250k
0.2615MSEHyperFM
Evaluation results
| Method | Links | |
|---|---|---|
| HyperFMTraining Setup=Full Fine-tuning, Model Capacity=32.06M, Trainable Param=32.06M, Train Time (Hr)=5.02@(2x48GB A6000)2026.04 | 0.2615 | |
| HyperFMTraining Setup=Decoder Only Fine-tuning, Model Capacity=32.06M, Trainable Param=1.48M, Train Time (Hr)=7.50@(2x48GB A6000)2026.04 | 0.3124 | |
| HyperSigmaTraining Setup=Decoder Only Fine-tuning, Model Capacity=100.16M, Trainable Param=0.69M, Train Time (Hr)=133.31@(2x12GB 2080Ti)2026.04 | 0.3212 | |
| CAMTraining Setup=Full Fine-tuning, Model Capacity=0.47M, Trainable Param=0.47M, Train Time (Hr)=3.71@(1x12GB RTX2080Ti)2026.04 | 0.3367 | |
| SpectralEarthTraining Setup=Full Fine-tuning, Model Capacity=88.78M, Trainable Param=88.78M, Train Time (Hr)=29.31@(4x24GB RTX6000)2026.04 | 0.3404 | |
| HyperFreeTraining Setup=Full Fine-tuning, Model Capacity=101.82M, Trainable Param=101.82M, Train Time (Hr)=3.09@(3x48GB RTX8000)2026.04 | 0.3523 | |
| CloudUNetTraining Setup=Full Fine-tuning, Model Capacity=1.86M, Trainable Param=1.86M, Train Time (Hr)=3.67@(1x12GB RTX2080Ti)2026.04 | 0.3752 | |
| UNetTraining Setup=Full Fine-tuning, Model Capacity=31.04M, Trainable Param=31.04M, Train Time (Hr)=1.57@(1x48GB A6000)2026.04 | 0.3928 | |
| HyperSigmaTraining Setup=Full Fine-tuning, Model Capacity=100.16M, Trainable Param=100.16M, Train Time (Hr)=4.54@(2x48GB A6000)2026.04 | 0.4649 | |
| SpectralEarthTraining Setup=Decoder Only Fine-tuning, Model Capacity=88.78M, Trainable Param=0.54M, Train Time (Hr)=9.58@(3x48GB RTX8000)2026.04 | 0.4699 | |
| HyperFreeTraining Setup=Decoder Only Fine-tuning, Model Capacity=101.82M, Trainable Param=0.69M, Train Time (Hr)=5.38@(2x48GB A6000)2026.04 | 0.557 | |
| HyperFreeTraining Setup=Zero-shot, Model Capacity=101.82M2026.04 | 9.17 | |
| HyperSigmaTraining Setup=Zero-shot, Model Capacity=100.16M2026.04 | 53.1539 | |
| SpectralEarthTraining Setup=Zero-shot, Model Capacity=88.78M2026.04 | 61.028 |