Skip to content

Experiments

This page summarizes qualitative experiments from the README, with the same scenarios preserved and clarified.

Experiment 1 (1-band corruption, 50%)

Preliminary sub-surface reconstruction with the single-band setup and 50% clustered occlusion after ~24h training.

Notes from the original run: - valid masks were used in training - land mask was used for visualization - loss was computed over the whole image - no inpainting-style known-pixel clamping in DDPM sampling - reported metrics: PSNR ~40 dB, SSIM ~0.90
img

Applying the same checkpoint to 75% occlusion still produced plausible reconstructions:
img

Experiment 2 (3-band x and y, EO condition, 95% corruption)

EO-conditioned multiband setup: - x: corrupted 3-band deeper input (depth levels 3,4,5) - y: clean 3-band deeper target - eo: additional surface condition

At 95% corruption, reconstructions remain strong, with noted risk of EO over-reliance.
img

Experiment 3 (High corruption and EO dropout)

Tested 75% corruption with 50% EO dropout while keeping the same checkpoint for comparison.

Observation: - removing EO condition reduces fidelity - outputs still match plausible distributional structure
img

After continued training (100 epochs in that setting), corruption increased to 95% while keeping 50% dropout.

Observation: - EO-present samples reconstruct faithfully - EO-absent samples remain realistic but can deviate from exact ground-truth patterns
img

Experiment 4 (High corruption, EO dropout, EO degradation)

To reduce direct EO value shortcutting, EO condition was degraded using random perturbations and speckle-like noise.

Observation: - reconstruction remains workable with degraded EO - model still performs plausible x -> y reconstruction when EO is absent or weak
img

Experiment 5 (Lower temperature levels)

Harder target depths were tested with levels: - 0 (EO) - 4 (5m) - 10 (16m) - 17 (50m)

Training setup: 95% corruption, 50% EO dropout, EO degradation.

Reported outcome: quality remained comparable on these depth levels.
img

Next Experiments (Roadmap)

  • Crank sparse corruption to mask_fraction=0.99 in EO multiband runs as the next baseline stress test.
  • Add more realistic corruption trajectories by replacing/augmenting random rectangles with "walk"-style masks that trace a path across each patch (submarine-like motion simulation).