Inferring high resolution imagery

1 . A method comprising:

obtaining image data capturing a particular geographic area during a particular time period, the image data including at least:

first image data captured in a first spectral band at a first spatial resolution, wherein the first image data depicts one or more terrain features of the particular geographic area, and

second image data captured in a second spectral band at a second spatial resolution less than the first spatial resolution;

applying, as input to a first upstream machine learning model, the first image data to generate, as output, one or more first spectral band embeddings, wherein the one or more first spectral band embeddings semantically represent the one or more terrain features of the particular geographic area;

applying, as input to a second upstream machine learning model, the second image data to generate, as output, one or more second spectral band embeddings; and

inferring second spectral band information about the one or more terrain features of the particular geographic area, wherein the inferring includes:

applying, as inputs to a downstream machine learning model, the one or more first spectral band embeddings and the one or more second spectral band embeddings to generate, as output, one or more inferred embeddings, wherein the one or more inferred embeddings indicate at least semantic information that corresponds to both the first spectral band and the second spectral band; and

generating, based on the one or more inferred embeddings, inferred image data, the inferred image data including image data in the second spectral band that captures the one or more terrain features at the first spatial resolution.

2 . The method of claim 1 , wherein the first image data and the second image data comprise high-elevation imagery captured by one or more satellites at separate times.

3 . The method of claim 1 , wherein the first image data and the second image data include corresponding temporal data including:

first temporal data associated with the first image data captured in the first spectral band at the first spatial resolution, wherein the first temporal data indicates one or more first particular times, during the particular time period, at which the first image data was captured, and

second temporal data associated with the second image data captured in the second spectral band at the second spatial resolution, wherein the second temporal data indicates one or more second particular times, during the particular time period, at which the second image data was captured,

wherein the one or more first particular times and the one or more second particular times occur at different times, and

wherein the one or more first spectral band embeddings and the one or more second spectral band embeddings include corresponding spatial and temporal information.

4 . The method of claim 3 , wherein the generated inferred image data in the second spectral band that captures the one or more terrain features of the particular geographic area at the first spatial resolution captures the one or more terrain features of the particular geographic area at the one or more first particular times.

5 . The method of claim 3 , wherein applying the first image data as input to the first upstream machine learning model and applying the second image data as input to the second upstream machine learning model includes applying the first image data and the second image data as corresponding time-series image data.

6 . The method of claim 1 , wherein the inferred image data is generated further based on upsampling the second image data to the first spatial resolution.

7 . The method of claim 1 , wherein the downstream machine learning model comprises a transformer machine learning model.

8 . The method of claim 1 , wherein applying the second image data as input to the second upstream machine learning model includes upsampling the second image data to the first spatial resolution and applying the upsampled second image data as input to the second upstream machine learning model.

9 . A system, comprising:

one or more processors; and

one or more memory storage devices storing instructions that, when executed by the one or more processors, cause the one or more processors to perform operations comprising:

obtaining image data capturing a particular geographic area during a particular time period, the image data including at least:

first image data captured in a first spectral band at a first spatial resolution, wherein the first image data depicts one or more terrain features of the particular geographic area, and

second image data captured in a second spectral band at a second spatial resolution less than the first spatial resolution;

applying, as input to a first upstream machine learning model, the first image data to generate, as output, one or more first spectral band embeddings, wherein the one or more first spectral band embeddings semantically represent the one or more terrain features of the particular geographic area;

applying, as input to a second upstream machine learning model, the second image data to generate, as output, one or more second spectral band embeddings; and

inferring second spectral band information about the one or more terrain features of the particular geographic area, wherein the inferring includes:

applying, as inputs to a downstream machine learning model, the one or more first spectral band embeddings and the one or more second spectral band embeddings to generate, as output, one or more inferred embeddings, wherein the one or more inferred embeddings indicate at least semantic information that corresponds to both the first spectral band and the second spectral band; and

generating, based on the one or more inferred embeddings, inferred image data, the inferred image data including image data in the second spectral band that captures the one or more terrain features at the first spatial resolution.

10 . The system of claim 9 , wherein the first image data and the second image data comprise high-elevation imagery captured by one or more satellites at separate times.

11 . The system of claim 9 , wherein the first image data and the second image data include corresponding temporal data including:

first temporal data associated with the first image data captured in the first spectral band at the first spatial resolution, wherein the first temporal data indicates one or more first particular times, during the particular time period, at which the first image data was captured, and

second temporal data associated with the second image data captured in the second spectral band at the second spatial resolution, wherein the second temporal data indicates one or more second particular times, during the particular time period, at which the second image data was captured,

wherein the one or more first particular times and the one or more second particular times occur at different times, and

wherein the one or more first spectral band embeddings and the one or more second spectral band embeddings include corresponding spatial and temporal information.

12 . The system of claim 11 , wherein the generated inferred image data in the second spectral band that captures the one or more terrain features of the particular geographic area at the first spatial resolution captures the one or more terrain features of the particular geographic area at the one or more first particular times.

13 . The system of claim 11 , wherein applying the first image data as input to the first upstream machine learning model and applying the second image data as input to the second upstream machine learning model includes applying the first image data and the second image data as corresponding time-series image data.

14 . The system of claim 9 , wherein the downstream machine learning model comprises a transformer machine learning model.

15 . The system of claim 9 , wherein applying the second image data as input to the second upstream machine learning model includes upsampling the second image data to the first spatial resolution and applying the upsampled second image data as input to the second upstream machine learning model.

16 . One or more non-transitory computer-readable storage mediums storing instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:

obtaining image data capturing a particular geographic area during a particular time period, the image data including at least:

first image data captured in a first spectral band at a first spatial resolution, wherein the first image data depicts one or more terrain features of the particular geographic area, and

second image data captured in a second spectral band at a second spatial resolution less than the first spatial resolution;

applying, as input to a first upstream machine learning model, the first image data to generate, as output, one or more first spectral band embeddings, wherein the one or more first spectral band embeddings semantically represent the one or more terrain features of the particular geographic area;

applying, as input to a second upstream machine learning model, the second image data to generate, as output, one or more second spectral band embeddings; and

inferring second spectral band information about the one or more terrain features of the particular geographic area, wherein the inferring includes:

applying, as inputs to a downstream machine learning model, the one or more first spectral band embeddings and the one or more second spectral band embeddings to generate, as output, one or more inferred embeddings, wherein the one or more inferred embeddings indicate at least semantic information that corresponds to both the first spectral band and the second spectral band; and

generating, based on the one or more inferred embeddings, inferred image data, the inferred image data including image data in the second spectral band that captures the one or more terrain features at the first spatial resolution.

17 . The one or more non-transitory computer-readable storage mediums of claim 16 , wherein the first image data and the second image data comprise high-elevation imagery captured by one or more satellites at separate times.

18 . The one or more non-transitory computer-readable storage mediums of claim 16 , wherein the first image data and the second image data include corresponding temporal data including:

first temporal data associated with the first image data captured in the first spectral band at the first spatial resolution, wherein the first temporal data indicates one or more first particular times, during the particular time period, at which the first image data was captured, and

second temporal data associated with the second image data captured in the second spectral band at the second spatial resolution, wherein the second temporal data indicates one or more second particular times, during the particular time period, at which the second image data was captured,

wherein the one or more first particular times and the one or more second particular times occur at different times, and

wherein the one or more first spectral band embeddings and the one or more second spectral band embeddings include corresponding spatial and temporal information.

19 . The one or more non-transitory computer-readable storage mediums of claim 18 , wherein the generated inferred image data in the second spectral band that captures the one or more terrain features of the particular geographic area at the first spatial resolution captures the one or more terrain features of the particular geographic area at the one or more first particular times.

20 . The one or more non-transitory computer-readable storage mediums of claim 18 , wherein applying the first image data as input to the first upstream machine learning model and applying the second image data as input to the second upstream machine learning model includes applying the first image data and the second image data as corresponding time-series image data.