Model Components

rslearn.models.simple_time_series.SimpleTimeSeries

SimpleTimeSeries is a feature extractor that wraps a unitemporal feature extractor and applies it on a time series. It encodes each image in the time series individually using the unitemporal feature extractor, and then pools the features temporally via max pooling, mean pooling, a ConvRNN, 3D convolutions, or 1D convolutions.

It inputs the time series as an image where the timesteps are stacked along the channel axis (as produced by a DataInput with multiple layers), and splits the stack back into individual images using image_keys. It outputs a FeatureMaps with the same structure as the wrapped feature extractor, but with the features pooled over time.

Configuration

Here is a summary, see rslearn.models.simple_time_series for all of the available options.

model:
  class_path: rslearn.train.lightning_module.RslearnLightningModule
  init_args:
    model:
      class_path: rslearn.models.multitask.MultiTaskModel
      init_args:
        encoder:
          - class_path: rslearn.models.simple_time_series.SimpleTimeSeries
            init_args:
              encoder:
                class_path: # ...
                init_args:
                  # ...
              # One of "max" (default), "mean", "convrnn", "conv3d", or
              # "conv1d".
              op: "max"
              # Number of layers for convrnn, conv3d, and conv1d ops.
              num_layers: null
              # A map from input dict keys to the number of bands per image.
              # This is used to split up the time series back into the
              # individual images.
              image_keys:
                sentinel2: 12
                sentinel1: 2
          - ...

Example

The main README has an example of using SimpleTimeSeries with SatlasPretrain, and the BitemporalSentinel2 example uses it for a bitemporal classification task.