Tessera

rslearn includes a wrapper for Tessera v1.1 QAT encoder checkpoints from https://github.com/ucam-eo/tessera.

The wrapper does not download weights automatically. We recommend using the Microsoft Planetary Computer checkpoint because rslearn does not currently support OPERA RTC-S1 here. Download the encoder-only tessera_v1_1_mpc_encoder.pt checkpoint from the official Tessera README, then pass it with checkpoint_path.

Tessera inputs should be normalized with rslearn.models.tessera.TesseraNormalize before they reach the model. Set the transform's data_source to match the checkpoint family:

• mpc: Microsoft Planetary Computer Sentinel-2 L2A and Sentinel-1 RTC.
• aws: AWS Open Data Earth-search Sentinel-2 L2A and ASF OPERA RTC-S1. This is not currently recommended because rslearn does not have an OPERA data source yet.

Tessera uses different normalization statistics for the two checkpoint families, which are selected by the data_source argument of TesseraNormalize.

Inputs

The model expects three raster inputs:

• s2: Sentinel-2 bands in Tessera order: B04, B02, B03, B08, B8A, B05, B06, B07, B11, B12.
• s1_ascending: Sentinel-1 ascending orbit bands vv, vh.
• s1_descending: Sentinel-1 descending orbit bands vv, vh.

All inputs must have timestamps. The wrapper uses each timestamp midpoint to compute day-of-year features.

TesseraNormalize expects Sentinel-1 values in standard power decibels (10 * log10(linear)). For raw linear RTC sources, apply Sentinel1ToDecibels before TesseraNormalize. If your Sentinel-1 layers are already stored in standard dB, skip Sentinel1ToDecibels and still run TesseraNormalize.

Example

Here is an example of a model config to compute embeddings with Tessera:

model:
  class_path: rslearn.train.lightning_module.RslearnLightningModule
  init_args:
    model:
      class_path: rslearn.models.singletask.SingleTaskModel
      init_args:
        encoder:
          - class_path: rslearn.models.tessera.Tessera
            init_args:
              # Replace with the path to the downloaded checkpoint!
              checkpoint_path: /path/to/tessera_v1_1_mpc_encoder.pt
              pixel_batch_size: 1024
        decoder:
          - class_path: rslearn.train.tasks.embedding.EmbeddingHead
    optimizer:
      class_path: rslearn.train.optimizer.AdamW
data:
  class_path: rslearn.train.data_module.RslearnDataModule
  init_args:
    inputs:
      s2:
        data_type: raster
        layers: ["sentinel2_l2a"]
        bands: ["B04", "B02", "B03", "B08", "B8A", "B05", "B06", "B07", "B11", "B12"]
        passthrough: true
        load_all_layers: true
      s1_ascending:
        data_type: raster
        layers: ["sentinel1_ascending"]
        bands: ["vv", "vh"]
        passthrough: true
        load_all_layers: true
      s1_descending:
        data_type: raster
        layers: ["sentinel1_descending"]
        bands: ["vv", "vh"]
        passthrough: true
        load_all_layers: true
    transforms:
      - class_path: rslearn.train.transforms.sentinel1.Sentinel1ToDecibels
        init_args:
          selectors: ["s1_ascending", "s1_descending"]
      - class_path: rslearn.models.tessera.TesseraNormalize
        init_args:
          data_source: mpc

The wrapper returns float32 feature maps with 192 channels by default, so it can be used with EmbeddingHead and RslearnWriter.

Data Source Example

Here is an example for obtaining Sentinel-2 L2A and Sentinel-1 RTC (with separated ascending and descending images) that are compatible with Tessera. It creates 12 chronological 30-day mosaics per layer, but note that Tessera can input more frequent images.

{
  "layers": {
    "sentinel2_l2a": {
      "type": "raster",
      "band_sets": [
        {
          "bands": [
            "B01",
            "B02",
            "B03",
            "B04",
            "B05",
            "B06",
            "B07",
            "B08",
            "B8A",
            "B09",
            "B11",
            "B12"
          ],
          "dtype": "uint16"
        }
      ],
      "data_source": {
        "class_path": "rslearn.data_sources.planetary_computer.Sentinel2",
        "init_args": {
          "cache_dir": "cache/planetary_computer",
          "harmonize": true,
          "sort_by": "eo:cloud_cover"
        },
        "ingest": false,
        "query_config": {
          "max_matches": 12,
          "min_matches": 12,
          "per_period_mosaic_reverse_time_order": false,
          "period_duration": "30d",
          "space_mode": "MOSAIC"
        }
      }
    },
    "sentinel1_ascending": {
      "type": "raster",
      "band_sets": [
        {
          "bands": ["vv", "vh"],
          "dtype": "float32"
        }
      ],
      "data_source": {
        "class_path": "rslearn.data_sources.planetary_computer.Sentinel1",
        "init_args": {
          "cache_dir": "cache/planetary_computer",
          "query": {
            "sar:instrument_mode": {
              "eq": "IW"
            },
            "sar:polarizations": {
              "eq": [
                "VV",
                "VH"
              ]
            },
            "sat:orbit_state": {
              "eq": "ascending"
            }
          }
        },
        "ingest": false,
        "query_config": {
          "max_matches": 12,
          "min_matches": 12,
          "per_period_mosaic_reverse_time_order": false,
          "period_duration": "30d",
          "space_mode": "MOSAIC"
        }
      }
    },
    "sentinel1_descending": {
      "type": "raster",
      "band_sets": [
        {
          "bands": ["vv", "vh"],
          "dtype": "float32"
        }
      ],
      "data_source": {
        "class_path": "rslearn.data_sources.planetary_computer.Sentinel1",
        "init_args": {
          "cache_dir": "cache/planetary_computer",
          "query": {
            "sar:instrument_mode": {
              "eq": "IW"
            },
            "sar:polarizations": {
              "eq": [
                "VV",
                "VH"
              ]
            },
            "sat:orbit_state": {
              "eq": "descending"
            }
          }
        },
        "ingest": false,
        "query_config": {
          "max_matches": 12,
          "min_matches": 12,
          "per_period_mosaic_reverse_time_order": false,
          "period_duration": "30d",
          "space_mode": "MOSAIC"
        }
      }
    }
  }
}