Research map / Point clouds / Shape and surface completion
Ice Cloud Geometry Retrieval with Calibrated Uncertainty from Passive Satellite Imagery
Ayush Prasad
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 |
|---|---|---|---|
| add ERA5 reanalysis conditioning (ablation) | G-QE · EarthCARE-VIIRS test | R2: about 8 points; without ERA5 R2 approx 0.66 | Improved |
| ensemble five quantile models (G-QE vs G-Q) (ablation) | ConvNextUNet quantile · EarthCARE-VIIRS test | R2: 0.731 -> 0.742 | Improved |
| ensemble five quantile models (G-QE vs G-Q) (ablation) | ConvNextUNet quantile · EarthCARE-VIIRS test | MPIW: 26.8 -> 25.1 | Improved |
| ensemble five quantile models (ablation) | ConvNextUNet quantile · thickness target | R2: .370 -> .407 | Improved |
| use quantile ensemble instead of deep ensemble of point models | G-DE · EarthCARE-VIIRS test | MPIW: 27.9 -> 25.1 | Improved |
| use quantile ensemble instead of deep ensemble | G-DE · EarthCARE-VIIRS test | Spearman rho: 0.242 -> 0.415 | Improved |
| add soft ordering loss for physics constraints (ablation) | G-Q · EarthCARE-VIIRS test | R2: 0.731 -> 0.712 | Hurt |
| add physics-constrained output head (ablation) | G-Q · EarthCARE-VIIRS test | R2: 0.731 -> 0.705 | Hurt |
| add post-hoc projection of ordering constraints (ablation) | G-Q · EarthCARE-VIIRS test | constraint violations (%): 0.293 -> 0 | Improved |
| add CQR calibration (ablation) | G-DE · EarthCARE-VIIRS test | coverage at 80% nominal (raw): 37.1% -> 90% coverage | Improved |
| replace quadratic ridge BT regression with G-QE | BT regression · EarthCARE-VIIRS test | R2: 0.590 -> 0.742 | Improved |
| add uncertainty features to cloud type classifier | geometry classifier · ISCCP cloud type | accuracy (%): 82.1 -> 84.9 | Improved |
Closest papers by meaning
- Cloud4D: Estimating Cloud Properties at a High Spatial and Temporal Resolution (NeurIPS 2025 · similarity 0.85)
- Generative Adversarial Training for Weakly Supervised Cloud Matting (ICCV 2019 · similarity 0.82)
- Morphing and Sampling Network for Dense Point Cloud Completion (AAAI 2020 · similarity 0.82)
- Flexible Depth Completion for Sparse and Varying Point Densities (CVPR 2024 · similarity 0.80)
- APML: Adaptive Probabilistic Matching Loss for Robust 3D Point Cloud Reconstruction (NeurIPS 2025 · similarity 0.80)
- Learnable Earth Parser: Discovering 3D Prototypes in Aerial Scans (CVPR 2024 · similarity 0.80)
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.
