levircd_dinov2/preprocessing.json

{
  "input_name": "input",
  "output_name": "logits",
  "input_shape": [
    "batch",
    2,
    3,
    448,
    448
  ],
  "input_layout": "(batch, 2 dates, 3 RGB channels, H, W)",
  "band_order": [
    "R",
    "G",
    "B"
  ],
  "value_range": "float32; divide 8-bit RGB by 255 BEFORE normalization",
  "normalization": {
    "mean": [
      0.485,
      0.456,
      0.406
    ],
    "std": [
      0.229,
      0.224,
      0.225
    ]
  },
  "input_size": 448,
  "dynamic_hw": false,
  "resize_to_input": "resize each 256px tile to 448px (bilinear) before inference; resize the 448px output mask back to display resolution",
  "tiling": {
    "tile_size": 256,
    "overlap": 0
  },
  "output": {
    "activation": "sigmoid",
    "meaning": "per-pixel change probability (channel 0)",
    "recommended_threshold": 0.5078125,
    "threshold_source": "val-selected (max-F1)"
  },
  "dinov2_note": "encoder runs on a FIXED 448px / 32x32 patch grid with interpolate_pos_encoding baked in; do NOT feed a different size."
}