ret, mask = cv2.threshold(mask, 1, 255, cv2.THRESH_BINARY)
# Denoise the mask
mask = SegmentationNetwork.denoise(mask)
# Invert the mask, since Emma painted the particles black rather than white in the mask layer,
# whereas the neural network uses classification label 0 for "not particle" and 1 for "particle"
mask = 255 - mask
# Open the file containing the RGB channel data
channels = mergetiff.rasterFromFile(channelsFile)
if channels.shape[2] > 3:
channels = channels[:, :, 0:3]
elif channels.shape[2] < 3:
Logger.error('could not extract RGB channel data!')
# Merge the RGB channels with the modified mask
shape = (channels.shape[0], channels.shape[1], 4)
merged = np.zeros(shape, dtype=channels.dtype)
merged[:, :, 0:3] = channels
merged[:, :, 3] = mask
# Write the output file (convert RGB to BGR for OpenCV)
if os.path.exists(outfile) == True:
Logger.warning('overwriting existing file {}'.format(outfile))
cv2.imwrite(outfile, merged[..., [2, 1, 0, 3]])
# Progress output
timer.end()
Logger.success('preprocessing complete ({}).'.format(timer.report()))
imageGroundTruth = '{}.groundtruth.png'.format(imageBase)
imagePrediction = '{}.prediction.png'.format(imageBase)
imageSliced = '{}.sliced.png'.format(imageBase)
cv2.imwrite(imageChannels, channels)
cv2.imwrite(imageGroundTruth, groundTruth)
cv2.imwrite(imagePrediction, prediction * 255)
cv2.imwrite(imageSliced, sliced)
# Generate the table row for our HTML report
tableRows.append(
'<tr><td><img src="{}"></td><td><img src="{}"></td><td><img src="{}"></td><td><img src="{}"></td><td>{}</td></tr>'
.format(
os.path.basename(imageChannels),
os.path.basename(imageGroundTruth),
os.path.basename(imagePrediction), os.path.basename(imageSliced),
''.join([
'<li><strong>Size:</strong> {:.0f} x {:.0f}, <strong>Ratio:</strong> {:.2f}</li>'
.format(particle['length'], particle['width'],
particle['ratio']) for particle in particleDetails
])))
# Save the HTML report
html = REPORT_TEMPLATE.replace('$$__ACCURACY__$$',
'{:.2f}%'.format(accuracy * 100.0))
html = html.replace('$$__ROWS__$$', '\n'.join(tableRows))
Utility.writeFile(os.path.join(outputDir, '_report.html'), html)
# Progress output
timer.end()
Logger.success('report generation complete ({}).'.format(timer.report()))
# Retrieve the list of input files
inputDir = Configuration.path('segmentation_data_preprocessed')
outputDir = Configuration.path('segmentation_data_sliced')
inputFiles = natsorted(glob.glob(os.path.join(inputDir, '*.png')))
# Progress output
print('Slicing preprocessed data for the segmentation network ({} files)...'.
format(len(inputFiles)))
# Keep track of processing progress and timing
numFiles = len(inputFiles)
timer = TimeLogger(numFiles, 'file')
progress = ProgressLogger(numFiles)
# Process each input file
for filenum, infile in enumerate(inputFiles):
# Progress output
progress.progress(filenum, 'Slicing input file "{}"...'.format(infile))
# Slice the file
SegmentationNetwork.sliceToDir(infile,
outputDir,
includeMask=True,
warnOnOverwrite=True)
# Progress output
timer.end()
Logger.success('slicing complete ({}).'.format(timer.report()))
