t.pytorch.predict - GRASS manual

NAME

t.pytorch.predict - Apply a pytorch model to imagery groups in a Space Time Raster Dataset (STRDS)

KEYWORDS

temporal, machine learning, deep learning, pytorch, unet, GPU, predict, imagery, raster, strds

SYNOPSIS

t.pytorch.predict

t.pytorch.predict --help

t.pytorch.predict [-cels] input=name [where=sql_query] [reference_strds=name[,name,...]] [reference_where=sql_query] [reference_suffix=string] [sampling=name[,name,...]] [offset=integer] [auxillary_group=name] [region_relation=string] output=name [title=string] [description=string] model=name model_code=name [vector_tiles=name] [tile_size=integer[,integer,...]] [overlap=integer] configuration=name [mask_json=name] [nprocs=integer] [basename=string] [--overwrite] [--help] [--verbose] [--quiet] [--ui]

Flags:

-c: Use CPU as device for prediction, default is use cuda (GPU) if detected
-e: Extend existing STRDS (requires overwrite flag)
-l: Limit output to valid range (data outside the valid range is set to valid min/max)
-s: Skip incomplete groups (do not fail)
--overwrite: Allow output files to overwrite existing files
--help: Print usage summary
--verbose: Verbose module output
--quiet: Quiet module output
--ui: Force launching GUI dialog

Parameters:

input=name [required]: Name of the input space time raster dataset
where=sql_query: WHERE conditions of SQL statement without 'where' keyword used in the temporal GIS framework; Example: start_time > '2001-01-01 12:30:00'
reference_strds=name[,name,...]: Name of the input space time raster datasets
reference_where=sql_query: WHERE conditions of SQL statement without 'where' keyword used in the temporal GIS framework; Where clause to select reference images
reference_suffix=string: Suffix to be added to the semantic label of the raster maps in the reference_strds
sampling=name[,name,...]: The method to be used for sampling the input dataset; Options: start, during, overlap, contain, equal, follows, precedes; Default: start
offset=integer: Offset that defines a reference map (e.g. -1 for the previous map (group) in the input STRDS)
auxillary_group=name: Input imagery group with time independent raster maps
region_relation=string: Process only maps with this spatial relation to the current computational region; Options: overlaps, contains, is_contained
output=name [required]: Name of the output space time raster dataset
title=string: Title of the resulting STRDS
description=string: Description of the resulting STRDS
model=name [required]: Path to input deep learning model file (.pt)
model_code=name [required]: Path to input deep learning model code (.py)
vector_tiles=name: Name of input vector map; Vector map with tiles to process (will be extended by "overlap")
tile_size=integer[,integer,...]: Number of rows and columns in tiles (rows, columns)
overlap=integer: Number of rows and columns of overlap in tiles
configuration=name [required]: Path to JSON file with band configuration in the input deep learning model
mask_json=name: JSON file with one or more mask band or map name(s) and reclass rules for masking, e.g. {"mask_band": "1 thru 12 36 = 1", "mask_map": "0"}
nprocs=integer: Number of threads for parallel computing; 0: use OpenMP default; >0: use nprocs; <0: use MAX-nprocs; Default: 0
basename=string: Name for output raster map

DESCRIPTION
EXAMPLES
- Run cloud detection on a Sentinel-3 SLSTR time series
- Estimate FSC on Sentinel-3 SLSTR time series
SEE ALSO
AUTHOR

DESCRIPTION

t.pytorch.predict is a wrapper around the i.pytorch.predict module and supports all relevant flags and options of that module. t.pytorch.predict compiles the input imagery groups to i.pytorch.predict from the temporal granules in the input STRDS. Those groups per granule are complemented with raster maps from a auxillary_group and/or another reference_strds, where maps in the reference STRDS are matched with the input STRDS in space and time using the user-defined sampling. If a reference STRDS or an auxaliry group is used it often makes sense to provide a basename for the resulting raster maps.

In order to run the module with tile- or orbit repeat-passes, the user should loop over tiles or orbits and use orbit- or tile IDs in the where clause of the input and reference STRDS. STRDS containing mosaics with equal spatial extent do not require special handling. Currently supported use-cases are:

only input STRDS, usually grouped ("one process per scene")
only input STRDS, usually grouped, with reference defined by offset (e.g. for "repeat-pass")
input STRDS and reference STRDS matched according to temporal relation given in sampling with single or grouped semantic labels

For more information on how machine learning models are applied to the imagery groups, please consult the manual of i.pytorch.predict.

EXAMPLES

Run cloud detection on a Sentinel-3 SLSTR time series

t.pytorch.predict -e --o --v input=Sentinel_3_SLSTR_RBT_L2 model=cloud.pt \
  output=Sentinel_3_SLSTR_RBT_L2 tile_size=1024,1024 overlap=164 \
  configuration=cloud.json model_code=S3_models/ nprocs=8 \
  mask_json=mask_land.json where="start_time > '2024-02-02'"

t.info Sentinel_3_SLSTR_RBT_L2
(...)

Estimate FSC on Sentinel-3 SLSTR time series

time t.pytorch.predict -e --o --v input=Sentinel_3_SLSTR_RBT_L2 model=fsc.pt \
  output=Sentinel_3_SLSTR_RBT_L2 tile_size=1024,1024 overlap=164 \
  configuration=fsc.json model_code=S3_models/ nprocs=8 \
  mask_json=mask_cloud_dl.json where="start_time > '2024-02-02'"

t.info Sentinel_3_SLSTR_RBT_L2
(...)

AUTHOR

Stefan Blumentrath, NVE

SOURCE CODE

Available at: t.pytorch.predict source code (history)

Accessed: Monday Jul 27 12:29:54 2026