Explainable and Resource-Aware Earth Observation AI
Real Sentinel-2 patches → pretrained cloud segmentation → measured features → symbolic rules → exact byte-budget selection. CPU-safe pretrained segmentation with optional NVIDIA CUDA and measured device comparisons.
0.526.8
0100
Mission preference
0100
01.5
Inference device
Changes to policy, budget, or selected images recalculate after a run. Newly selected images require real inference; cached images are labeled. Auto selects CUDA only when usable; otherwise inference stays on CPU. Budget bytes are the measured compressed B2/B3/B4/B8 image payload; ground-truth labels are excluded from transmitted-image size.
Adjust the policy and run the pipeline to measure a mission.
Ready. Official pretrained weights load on first CPU/GPU inference; the dedicated ZeroGPU Space preloads CUDA weights during startup.