feature: save labeled segmentations rather than binary segmentations

IFTPipeline.jl/Project.toml

@@ -8,9 +8,11 @@ ArgParse = "c7e460c6-2fb9-53a9-8c5b-16f535851c63"
 CSV = "336ed68f-0bac-5ca0-87d4-7b16caf5d00b"
 CSVFiles = "5d742f6a-9f54-50ce-8119-2520741973ca"
 FileIO = "5789e2e9-d7fb-5bc7-8068-2c6fae9b9549"
+FixedPointNumbers = "53c48c17-4a7d-5ca2-90c5-79b7896eea93"
 Folds = "41a02a25-b8f0-4f67-bc48-60067656b558"
 HDF5 = "f67ccb44-e63f-5c2f-98bd-6dc0ccc4ba2f"
 IceFloeTracker = "04643c7a-9ac6-48c5-822f-2704f9e70bd3"
+ImageSegmentation = "80713f31-8817-5129-9cf8-209ff8fb23e1"
 Images = "916415d5-f1e6-5110-898d-aaa5f9f070e0"
 Logging = "56ddb016-857b-54e1-b83d-db4d58db5568"
 LoggingExtras = "e6f89c97-d47a-5376-807f-9c37f3926c36"

IFTPipeline.jl/src/IFTPipeline.jl

@@ -16,6 +16,8 @@ using Pkg
 using FileIO
 using Images
 using CSV
+using ImageSegmentation
+using FixedPointNumbers
 
 include("cli.jl")
 include("soit-parser.jl")

IFTPipeline.jl/src/cli.jl

@@ -75,7 +75,7 @@ function mkcliextract!(settings)
 
         "--features", "-f"
         help = """Features to extract. Format: "feature1 feature2". For an extensive list of extractable features see https://scikit-image.org/docs/stable/api/skimage.measure.html#skimage.measure.regionprops:~:text=The%20following%20properties%20can%20be%20accessed%20as%20attributes%20or%20keys"""
-        default = "centroid area major_axis_length minor_axis_length convex_area bbox orientation perimeter"
+        default = "label centroid area major_axis_length minor_axis_length convex_area bbox orientation perimeter"
     end
     return nothing
 end
@@ -104,7 +104,7 @@ function mkcliextract_single!(settings)
         help = """Features to extract. Format: "feature1 feature2". For an extensive list of extractable features see https://scikit-image.org/docs/stable/api/skimage.measure.html#skimage.measure.regionprops:~:text=The%20following%20properties%20can%20be%20accessed%20as%20attributes%20or%20keys"""
         nargs = '+'
         arg_type = String
-        default = ["centroid", "area", "major_axis_length", "minor_axis_length", "convex_area", "bbox", "orientation", "perimeter"]
+        default = ["label", "centroid", "area", "major_axis_length", "minor_axis_length", "convex_area", "bbox", "orientation", "perimeter"]
     end
     return nothing
 end

IFTPipeline.jl/src/feature-extraction.jl

@@ -48,16 +48,16 @@ julia> IFTPipeline.extractfeatures(bw_img; minarea=minarea, maxarea=maxarea, fea
 ```
 """
 function extractfeatures(
-    bw::T;
+    labeled_floes::AbstractArray{<:Integer};
     minarea::Int64=350,
     maxarea::Int64=90000,
     features::Union{Vector{Symbol},Vector{<:AbstractString}}
-)::DataFrame where {T<:AbstractArray{Bool}}
+)::DataFrame
     # assert the first area threshold is less than the second
     minarea >= maxarea &&
         throw(ArgumentError("The minimum area must be less than the maximum area."))
 
-    props = regionprops_table(label_components(bw, trues(3, 3)); properties=features)
+    props = regionprops_table(labeled_floes; properties=features)
 
     # filter by area using the area thresholds
     return props[minarea.<=props.area.<=maxarea, :]
@@ -94,11 +94,12 @@ end
 function extractfeatures_single(;
     input::String, output::String, minarea::Int64, maxarea::Int64, features::Array{String}
 )
-    @info "Loading segmented floes as a binarized image from $input"
-    segmented_floes = BitMatrix(FileIO.load(input))
+    @info "Loading segmented floes from $input"
+    image = FileIO.load(input)
+    labeled_floes = Int64.(reinterpret.(real.(image)))
 
     @info "Extracting features from each floe: $features"
-    props = IFTPipeline.extractfeatures(segmented_floes; minarea=minarea, maxarea=maxarea, features=features)
+    props = IFTPipeline.extractfeatures(labeled_floes; minarea=minarea, maxarea=maxarea, features=features)
 
     @info "Extracted properties:"
     @info props

IFTPipeline.jl/src/preprocess.jl

@@ -287,10 +287,15 @@ function preprocess_single(; truecolor::T, falsecolor::T, landmask::T, landmask_
 
     @info "Segmenting floes started"
     segmented_floes = preprocess(rgb_truecolor_img, rgb_falsecolor_img, landmask)
-    @info "Segmenting floes complete"
+    @debug "Segmenting floes complete: $segmented_floes"
+
+    @info "Labeling floes"
+    labeled_floes = label_components(segmented_floes)
+    labeled_floes_cast = convert(Array{choose_dtype(maximum(labeled_floes))}, labeled_floes)
+    @debug "Labeled floes: $labeled_floes"
 
     @info "Writing segmented floes to $output"
-    FileIO.save(output, segmented_floes)
+    FileIO.save(output, labeled_floes_cast)
 
     return nothing
 end

IFTPipeline.jl/src/tracker.jl

@@ -89,7 +89,7 @@ function track_single(;
 
     # Load the files – can we drop the memory requirements by doing two at once?
     @info "Loading $imgs"
-    imgs_ = [Integer.(rawview(channelview((FileIO.load(img))))) for img in imgs]
+    imgs_ = [Int64.(rawview(channelview((FileIO.load(img))))) for img in imgs]
 
     @info "Loading $props"
     props_ = [DataFrame(CSV.File(prop)) for prop in props]

test/test-IFTPipeline.jl-cli.sh

@@ -8,7 +8,7 @@
 # Refs:
 # - https://stackoverflow.com/a/28085062/24937841 and 
 # - https://unix.stackexchange.com/a/31712
-: "${IFT:=julia --project=../IFTPipeline.jl ../IFTPipeline.jl/src/cli.jl}"
+: "${IFT:=julia --project=`pwd`/../IFTPipeline.jl `pwd`/../IFTPipeline.jl/src/cli.jl}"
 echo "IFT=${IFT}"
 
 # Data target
@@ -21,32 +21,34 @@ echo "DATA_TARGET=${DATA_TARGET}"
 echo "DATA_SOURCE=${DATA_SOURCE}"
 
 cp -r ${DATA_SOURCE}/* ${DATA_TARGET}/
-SAMPLEIMG=${DATA_TARGET}/20220914.terra.truecolor.250m.tiff
+cd ${DATA_TARGET}
+echo "in ${DATA_TARGET}"
+
+SAMPLEIMG=20220914.terra.truecolor.250m.tiff
 
 # Set up debug messages
 export JULIA_DEBUG="Main,IFTPipeline,IceFloeTracker" 
 
 # Run the processing (single files)
-LANDMASK=${DATA_TARGET}/landmask.tiff
-LANDMASK_NON_DILATED=${DATA_TARGET}/landmask.non-dilated.tiff
-LANDMASK_DILATED=${DATA_TARGET}/landmask.dilated.tiff
+LANDMASK=landmask.tiff
+LANDMASK_NON_DILATED=landmask.non-dilated.tiff
+LANDMASK_DILATED=landmask.dilated.tiff
+
 ${IFT} landmask_single -i ${LANDMASK} -o ${LANDMASK_NON_DILATED} -d ${LANDMASK_DILATED}
 
 for satellite in "aqua" "terra"
 do
-    TRUECOLOR=${DATA_TARGET}/20220914.${satellite}.truecolor.250m.tiff
-    FALSECOLOR=${DATA_TARGET}/20220914.${satellite}.falsecolor.250m.tiff
-    SEGMENTED=${DATA_TARGET}/20220914.${satellite}.segmented.250m.tiff
-    FLOEPROPERTIES=${DATA_TARGET}/20220914.${satellite}.segmented.250m.props.csv
-    HDF5FILE=${DATA_TARGET}/20220914.${satellite}.h5
-    ${IFT} preprocess_single --truecolor ${TRUECOLOR} --falsecolor ${FALSECOLOR} --landmask ${LANDMASK_NON_DILATED} --landmask-dilated ${LANDMASK_DILATED} --output ${SEGMENTED}
-    ${IFT} extractfeatures_single --input ${SEGMENTED} --output ${FLOEPROPERTIES}
-    ${IFT} makeh5files_single --passtime "2022-09-14T12:00:00" --truecolor ${TRUECOLOR} --falsecolor ${FALSECOLOR} --labeled ${SEGMENTED} --props ${FLOEPROPERTIES} --output ${HDF5FILE}
+    TRUECOLOR=20220914.${satellite}.truecolor.250m.tiff
+    FALSECOLOR=20220914.${satellite}.falsecolor.250m.tiff
+    LABELED=20220914.${satellite}.labeled.250m.tiff
+    FLOEPROPERTIES=20220914.${satellite}.labeled.250m.props.csv
+    HDF5FILE=20220914.${satellite}.h5
+    ${IFT} preprocess_single --truecolor ${TRUECOLOR} --falsecolor ${FALSECOLOR} --landmask ${LANDMASK_NON_DILATED} --landmask-dilated ${LANDMASK_DILATED} --output ${LABELED}
+    ${IFT} extractfeatures_single --input ${LABELED} --output ${FLOEPROPERTIES}
+    ${IFT} makeh5files_single --passtime "2022-09-14T12:00:00" --truecolor ${TRUECOLOR} --falsecolor ${FALSECOLOR} --labeled ${LABELED} --props ${FLOEPROPERTIES} --output ${HDF5FILE}
 done
 
-${IFT} track_single --imgs ${DATA_TARGET}/20220914.{aqua,terra}.segmented.250m.tiff --props ${DATA_TARGET}/20220914.{aqua,terra}.segmented.250m.props.csv --latlon ${TRUECOLOR} --passtimes "2022-09-14T12:00:00" "2022-09-15T12:00:00" --output ${DATA_TARGET}/paired-floes.csv
-
-
+${IFT} track_single --imgs 20220914.{aqua,terra}.labeled.250m.tiff --props 20220914.{aqua,terra}.labeled.250m.props.csv --latlon ${TRUECOLOR} --passtimes "2022-09-14T12:00:00" "2022-09-15T12:00:00" --output paired-floes.csv
 
 # Run the processing (batch)
 ${IFT} landmask ${DATA_TARGET} ${DATA_TARGET}