def __init__(self, parent=None):
super(MyMainWindow, self).__init__(parent)
qApp.installEventFilter(self)
self.setupUi(self)
self.data = ImageData()
self.errorMsg = QErrorMessage(self)
self.errorMsg.setWindowModality(Qt.WindowModal)
self.pushButtonLoadWL.clicked.connect(self.loadWhite)
self.pushButtonLoadSpectra.clicked.connect(self.loadSpectra)
self.pushButtonLoad.clicked.connect(self.loadAnnotations)
self.pushButtonSave.clicked.connect(self.saveAnnotations)
self.pushButtonClear.clicked.connect(self.plot_visual.clearSelected)
# self.comboBoxMethod.currentIndexChanged.connect()
self.horizontalSliderWn.valueChanged.connect(self.wavenumberSlide)
self.lineEditWn.editingFinished.connect(self.wavenumberEdit)
self.comboBoxVisual.currentIndexChanged.connect(self.imageProjection)
self.comboBoxCmaps.currentTextChanged.connect(self.plot_visual.setCmap)
self.lineEditWn.setFormat("%.2f")
# self.lineEditWavenumber.setRange(min=wmin, max=wmax, default=.5*(wmin+wmax))
self.comboBoxAnnotation.currentIndexChanged.connect(
self.plot_visual.setAnnotation)
self.plot_visual.addedPoint.connect(self.plot_whitelight.addPoint)
self.plot_visual.removedPoint.connect(self.plot_whitelight.removePoint)
self.plot_visual.alteredCounts.connect(self.updateCounts)
self.plot_whitelight.clickedPoint.connect(
self.plot_visual.clickOnePoint)
def run_instance(self, category_path, category):
print 'Estimating volume for path', category_path
try:
image_data = ImageData(category_path)
instance = category_path.split('/')[-2] if category_path.endswith(
'/') else category_path.split('/')[-1]
self.stats.add_true_for_instance(category, instance,
image_data.volume['value'])
volume = self.volume_estimator.with_kmeans().estimate(image_data)
self.stats.add_predicted_for_instance(
category, instance, self.volume_estimator.method.desc, volume)
if debug():
print u'Predicted volume for target %s = %3.9f %s\u00B3 with %s.' % (
instance, volume, image_data.volume['units'],
self.volume_estimator.method.desc)
volume = self.volume_estimator.with_meanshift().estimate(
image_data)
self.stats.add_predicted_for_instance(
category, instance, self.volume_estimator.method.desc, volume)
if debug():
print u'Predicted volume for target %s = %3.9f %s\u00B3 with %s.' % (
instance, volume, image_data.volume['units'],
self.volume_estimator.method.desc)
except Exception as e:
print "Exception occurred dring volume estimation, skipping image at", category_path, e
def scrape(self):
# Get the HTML from a random page.
page_offset = str(random.randint(0, self.UPPER_BOUND))
html_doc = requests.get(self.URL + "?offset=" + page_offset).text
# Create the soup object for extracting tags and stuff.
soup = BeautifulSoup(html_doc)
# Find all the thumbnail images.
thumbnails = soup.find_all("a", {"class": self.THUMBNAIL_CLASS})
image_data_items = []
for thumbnail in thumbnails:
if not thumbnail.img.has_attr("alt"):
continue
descr = thumbnail.img["alt"]
# if thumbnail.has_attr(self.FULL_IMG_ATTR):
# image_data_items.append(ImageData(thumbnail[self.FULL_IMG_ATTR], descr))
if thumbnail.has_attr(self.IMG_ATTR):
image_data_items.append(
ImageData(thumbnail[self.IMG_ATTR], descr))
else:
image_data_items.append(ImageData(thumbnail.img["src"], descr))
return image_data_items
def setupThreads(self):
# Shared data to connect all image services
self.imData = ImageData(FalconEyeMap.VID_1)
# Retireve image thread
self.recvImg = RetrieveImage()
self.recvImg.setSharedData(self.imData)
# Processing image thread
self.imgProcess = ProcessImage()
self.imgProcess.setSharedData(self.imData)
# Show the video window (Debbugging only)
self.debugFeed = DebugVideoFeed()
self.debugFeed.setSharedData(self.imData)
# 'plan' for TensorRT or 'onnx' for ONNX
backend = 'plan'
retina_backends = {'onnx': RetinaInferORT, 'plan': RetinaInferTRT}
model_dir, model_path = config.build_model_paths(model_name, backend)
retina_backend = retina_backends[backend](rec_name=model_path)
detector = FaceDetector(inference_backend=retina_backend)
detector.prepare(fix_image_size=(input_shape[1], input_shape[0]))
tt0 = time.time()
image = cv2.imread('test_images/Stallone.jpg.jpg', cv2.IMREAD_COLOR)
image = ImageData(image, input_shape)
image.resize_image(mode='pad')
tt1 = time.time()
print(f"Preparing image took: {tt1 - tt0}")
t0 = time.time()
for i in range(iters):
tf0 = time.time()
detections = detector.detect(image.transformed_image, threshold=0.3)
tf1 = time.time()
print(f"Full detection took: {tf1 - tf0}")
t1 = time.time()
print(f'Took {t1 - t0} s. ({iters / (t1 - t0)} im/sec)')
class MyMainWindow(QMainWindow, Ui_MainWindow):
closing = pyqtSignal()
fileOptions = QFileDialog.Options() | QFileDialog.DontUseNativeDialog
def __init__(self, parent=None):
super(MyMainWindow, self).__init__(parent)
qApp.installEventFilter(self)
self.setupUi(self)
self.data = ImageData()
self.errorMsg = QErrorMessage(self)
self.errorMsg.setWindowModality(Qt.WindowModal)
self.pushButtonLoadWL.clicked.connect(self.loadWhite)
self.pushButtonLoadSpectra.clicked.connect(self.loadSpectra)
self.pushButtonLoad.clicked.connect(self.loadAnnotations)
self.pushButtonSave.clicked.connect(self.saveAnnotations)
self.pushButtonClear.clicked.connect(self.plot_visual.clearSelected)
# self.comboBoxMethod.currentIndexChanged.connect()
self.horizontalSliderWn.valueChanged.connect(self.wavenumberSlide)
self.lineEditWn.editingFinished.connect(self.wavenumberEdit)
self.comboBoxVisual.currentIndexChanged.connect(self.imageProjection)
self.comboBoxCmaps.currentTextChanged.connect(self.plot_visual.setCmap)
self.lineEditWn.setFormat("%.2f")
# self.lineEditWavenumber.setRange(min=wmin, max=wmax, default=.5*(wmin+wmax))
self.comboBoxAnnotation.currentIndexChanged.connect(
self.plot_visual.setAnnotation)
self.plot_visual.addedPoint.connect(self.plot_whitelight.addPoint)
self.plot_visual.removedPoint.connect(self.plot_whitelight.removePoint)
self.plot_visual.alteredCounts.connect(self.updateCounts)
self.plot_whitelight.clickedPoint.connect(
self.plot_visual.clickOnePoint)
def closeEvent(self, event):
self.closing.emit()
plt.close('all')
self.deleteLater()
qApp.quit()
def loadSpectra(self):
"Load a file or return (error message, traceback) from trying"
fileName, _ = QFileDialog.getOpenFileName(
self,
"Load FTIR image",
"",
"Matlab files (*.mat);;All files (*)",
options=MyMainWindow.fileOptions)
self.data.readmat(fileName)
self.plot_visual.setData(self.data.wavenumber, self.data.raw,
self.data.wh)
self.horizontalSliderWn.setMaximum(len(self.data.wavenumber) - 1)
self.lineEditWn.setRange(self.data.wmin, self.data.wmax)
self.imageProjection()
self.plot_whitelight.load(os.path.splitext(fileName)[0] + '.jpg')
def loadAnnotations(self):
fileName, _ = QFileDialog.getOpenFileName(
self,
"Load annotation map",
"",
"Matlab files (*.mat);;All files (*)",
options=MyMainWindow.fileOptions)
if fileName:
s = scipy.io.loadmat(fileName)['annotations']
self.plot_visual.setAnnotations(s)
def saveAnnotations(self):
fileName, _ = QFileDialog.getSaveFileName(
self,
"Save annotation map",
"",
"Matlab files (*.mat);;All files (*)",
options=MyMainWindow.fileOptions)
if fileName:
scipy.io.savemat(
fileName, {'annotations': self.plot_visual.getAnnotations()})
# Image visualization
def loadWhite(self):
fileName, _ = QFileDialog.getOpenFileName(
self,
"Load white light image",
"",
"Image files (*.jpg *.png);;All files (*)",
options=MyMainWindow.fileOptions)
if fileName:
self.plot_whitelight.load(fileName)
def imageProjection(self):
meth = self.comboBoxVisual.currentIndex()
iswn = meth == 2
self.lineEditWn.setEnabled(iswn)
self.horizontalSliderWn.setEnabled(iswn)
self.plot_visual.setProjection(meth,
-1 - self.horizontalSliderWn.value())
def updateCounts(self, counts):
for i in range(self.comboBoxAnnotation.count()):
t = self.comboBoxAnnotation.itemText(i)
p = t.rfind('(')
t = t[:p] if p >= 0 else t + ' '
self.comboBoxAnnotation.setItemText(i, t + '(%d)' % counts[i])
def sliderToBox(self, slider, box, ixfinder):
wn = self.plot_visual.getWavenumbers()
if wn is not None:
if (not box.hasAcceptableInput()
or slider.value() != ixfinder(wn, box.value())):
box.setValue(wn[-1 - slider.value()])
elif (not box.hasAcceptableInput() or slider.value() != int(
round(slider.maximum() * (box.value() - self.wmin) /
(self.wmax - self.wmin)))):
box.setValue(self.wmin + (self.wmax - self.wmin) * slider.value() /
slider.maximum())
def boxToSlider(self, slider, box, ixfinder):
wn = self.plot_visual.getWavenumbers()
if wn is not None:
if box.hasAcceptableInput():
slider.setValue(ixfinder(wn, box.value()))
else:
box.setValue(wn[-1 - slider.value()])
else:
slider.setValue(
int(
round(slider.maximum() * (box.value() - self.wmin) /
(self.wmax - self.wmin))))
def wavenumberSlide(self):
self.sliderToBox(
self.horizontalSliderWn, self.lineEditWn,
lambda wn, val: len(wn) - 1 - (np.abs(wn - val)).argmin())
self.plot_visual.setProjection(2, -1 - self.horizontalSliderWn.value())
def wavenumberEdit(self):
self.boxToSlider(
self.horizontalSliderWn, self.lineEditWn,
lambda wn, val: len(wn) - 1 - (np.abs(wn - val)).argmin())
def set_image_data(self, image, um_per_pixel):
self.image_data = ImageData(image.index, um_per_pixel, image)
log.info("Image Index: %s" % image.index)
class FastqImageAligner(object):
"""A class to find the alignment of fastq data and image data."""
def __init__(self, microns_per_pixel):
self.fastq_tiles = {}
self.fastq_tiles_keys = []
self.microns_per_pixel = microns_per_pixel
self.image_data = None
self.fq_w = 935 #um
self.control_corr = 0
self.non_mutual_hits = set()
self.mutual_hits = set()
self.bad_mutual_hits = set()
self.good_mutual_hits = set()
self.exclusive_hits = set()
self.hitting_tiles = []
def load_reads(self, tile_data, valid_keys=None):
start = time.time()
for tile_key, read_names in tile_data.items():
if valid_keys is None or tile_key in valid_keys:
self.fastq_tiles[tile_key] = FastqTileRCs(
tile_key, read_names, self.microns_per_pixel)
@property
def fastq_tiles_list(self):
for _, tile in sorted(self.fastq_tiles.items()):
yield tile
def all_reads_fic_from_aligned_fic(self, other_fic, all_reads):
self.load_reads(
all_reads,
valid_keys=[tile.key for tile in other_fic.hitting_tiles])
self.image_data = deepcopy(other_fic.image_data)
self.fq_w = other_fic.fq_w
self.set_fastq_tile_mappings()
self.set_all_fastq_image_data()
self.hitting_tiles = [
self.fastq_tiles[tile.key] for tile in other_fic.hitting_tiles
]
self.clusters = other_fic.clusters
for other_tile in other_fic.hitting_tiles:
tile = self.fastq_tiles[other_tile.key]
try:
tile.set_aligned_rcs_given_transform(other_tile.scale,
other_tile.rotation,
other_tile.offset)
except AttributeError:
# I'm not sure why this is happening - tiles aren't getting aligned?
# We will make sure tiles are aligned before writing results to disk
log.debug("Skipping tile that lacks rotation!")
continue
def set_tile_alignment(self, tile_key, scale, fq_w, rotation, rc_offset):
if self.fastq_tiles[tile_key] not in self.hitting_tiles:
self.hitting_tiles.append(self.fastq_tiles[tile_key])
self.fq_w = fq_w
self.set_fastq_tile_mappings()
self.set_all_fastq_image_data()
tile = self.fastq_tiles[tile_key]
tile.set_aligned_rcs_given_transform(scale, rotation, rc_offset)
def alignment_from_alignment_file(self, path):
self.hitting_tiles = []
with open(path) as f:
astats = stats.AlignmentStats().from_file(f)
for tile_key, scaling, tile_width, rotation, rc_offset, _ in astats:
self.set_tile_alignment(tile_key, scaling, tile_width, rotation,
rc_offset)
def set_sexcat_from_file(self, fpath, cluster_strategy):
with open(fpath) as f:
#FLETCHER TIMING start = time.time()
self.clusters = clusters.Clusters(f, cluster_strategy)
#FLETCHER log.info("Clusters File: %s " % self.clusters.point_rcs)
def set_image_data(self, image, um_per_pixel):
self.image_data = ImageData(image.index, um_per_pixel, image)
log.info("Image Index: %s" % image.index)
def set_all_fastq_image_data(self):
for key, tile in self.fastq_tiles.items():
tile.set_fastq_image_data(self.fq_im_offset, self.fq_im_scale,
self.fq_im_scaled_dims, self.fq_w)
def rotate_all_fastq_data(self, degrees):
im_shapes = [
tile.rotate_data(degrees) for tile in self.fastq_tiles_list
]
self.fq_im_scaled_dims = np.array(im_shapes).max(axis=0)
for tile in self.fastq_tiles_list:
tile.image_shape = self.fq_im_scaled_dims
def set_fastq_tile_mappings(self):
"""Calculate parameters for mapping fastq tiles for ffts."""
log.info("Setting Fastq Tile Mappings")
assert self.image_data is not None, 'No image data loaded.'
assert self.fastq_tiles != {}, 'No fastq data loaded.'
all_data = np.concatenate(
[tile.rcs for key, tile in self.fastq_tiles.items()])
x_min, y_min = all_data.min(axis=0)
x_max, y_max = all_data.max(axis=0)
self.fq_im_offset = np.array([-x_min, -y_min])
log.info("Fastq Image Offset: %s" % self.fq_im_offset)
log.info("x_min, y_min: %s, %s" % (x_min, y_min))
self.fq_im_scale = (float(self.fq_w) /
(x_max - x_min)) / self.image_data.um_per_pixel
log.info('Fastq Image Scale: %s' % self.fq_im_scale)
self.fq_im_scaled_maxes = self.fq_im_scale * np.array(
[x_max - x_min, y_max - y_min])
self.fq_im_scaled_dims = (self.fq_im_scaled_maxes + [1, 1]).astype(
np.int)
def find_hitting_tiles(self,
possible_tile_keys,
cluster_size,
snr_thresh=1.2):
log.info('Locating Hitting Tiles')
possible_tiles = [
self.fastq_tiles[key] for key in possible_tile_keys
if key in self.fastq_tiles
]
#FLETCHER log.debug("Possible Tiles: %s" % (possible_tile_keys))
impossible_tiles = [
tile for tile in self.fastq_tiles.values()
if tile not in possible_tiles
]
impossible_tiles.sort(key=lambda tile: -len(tile.read_names))
log.info("Sorting Impossible Tiles Method: %s" %
(-len(tile.read_names)))
control_tiles = impossible_tiles[:2]
#FLETCHER TIMING
# start = time.time()
self.image_data.set_fft(self.fq_im_scaled_dims)
self.control_corr = 0
start_control = time.time()
for control_tile in control_tiles:
log.info("Starting Control Correlation")
corr, _ = control_tile.fft_align_with_im(self.image_data,
cluster_size)
if corr > self.control_corr:
self.control_corr = corr
#FLETCHER log.info('Control Correlation: %s seconds' % round(time.time()- start_control))
del control_tiles
self.hitting_tiles = []
log.info("Starting Experimental Correlation")
for tile in possible_tiles:
start = time.time()
log.info("Trying Tile: %s" % tile.key.title())
max_corr, align_tr = tile.fft_align_with_im(
self.image_data, cluster_size)
log.info("Correlations: Max: %s Control: %s" %
(max_corr, self.control_corr))
log.info("FM Cross-correllation Time: %s second" %
round(time.time() - start, 0))
#FLETCHER log.debug("Image Data: %s" % self.image_data.um_dims)
if max_corr > snr_thresh * self.control_corr:
tile.set_aligned_rcs(align_tr)
self.max_correlation = max_corr
log.info("Tile Max Correlation: %s" % self.max_correlation)
tile.snr = max_corr / self.control_corr
self.hitting_tiles.append(tile)
def find_points_in_frame(self, consider_tiles='all'):
self.rcs_in_frame = []
aligned_rcs_in_frame = []
im_shape = self.image_data.image.shape
if consider_tiles == 'all':
considered_tiles = self.hitting_tiles
else:
considered_tiles = [consider_tiles]
for tile in considered_tiles:
rcs = tile.rcs.astype(np.int)
for i, pt in enumerate(tile.aligned_rcs):
if 0 <= pt[0] < im_shape[0] and 0 <= pt[1] < im_shape[1]:
aligned_rcs_in_frame.append(pt)
self.rcs_in_frame.append((tile.key, rcs[i]))
self.aligned_rcs_in_frame = np.array(aligned_rcs_in_frame)
def hit_dists(self, hits):
return [self.single_hit_dist(hit) for hit in hits]
def single_hit_dist(self, hit):
return np.linalg.norm(self.clusters.point_rcs[hit[0]] -
self.aligned_rcs_in_frame[hit[1]])
def remove_longest_hits(self, hits, pct_thresh):
if not hits:
return []
dists = self.hit_dists(hits)
thresh = np.percentile(dists, pct_thresh * 100)
return [hit for hit in hits if self.single_hit_dist(hit) <= thresh]
def find_hits(self, consider_tiles='all'):
# --------------------------------------------------------------------------------
# Find nearest neighbors
# --------------------------------------------------------------------------------
self.find_points_in_frame(consider_tiles)
cluster_tree = KDTree(self.clusters.point_rcs)
aligned_tree = KDTree(self.aligned_rcs_in_frame)
# All indices are in the order (cluster_index, aligned_in_frame_idx)
cluster_to_aligned_indexes = set()
for i, pt in enumerate(self.clusters.point_rcs):
dist, idx = aligned_tree.query(pt)
cluster_to_aligned_indexes.add((i, idx))
aligned_to_cluster_indexs_rev = set()
for i, pt in enumerate(self.aligned_rcs_in_frame):
dist, idx = cluster_tree.query(pt)
aligned_to_cluster_indexs_rev.add((idx, i))
# --------------------------------------------------------------------------------
# Find categories of hits
# --------------------------------------------------------------------------------
mutual_hits = cluster_to_aligned_indexes & aligned_to_cluster_indexs_rev
non_mutual_hits = cluster_to_aligned_indexes ^ aligned_to_cluster_indexs_rev
cluster_in_non_mutual = set(i for i, j in non_mutual_hits)
aligned_in_non_mutual = set(j for i, j in non_mutual_hits)
exclusive_hits = set((i, j) for i, j in mutual_hits
if i not in cluster_in_non_mutual
and j not in aligned_in_non_mutual)
# --------------------------------------------------------------------------------
# Recover good non-exclusive mutual hits.
# --------------------------------------------------------------------------------
# If the distance to second neighbor is too close, that suggests a bad peak call combining
# two peaks into one. Filter those out with a gaussian-mixture-model-determined threshold.
if int(16.0 / self.image_data.um_per_pixel) == 60:
# Value decided by observation of our data. May vary with equipment.
good_hit_threshold = 5
else:
good_hit_threshold = np.percentile(self.hit_dists(exclusive_hits),
95)
second_neighbor_thresh = 2 * good_hit_threshold
exclusive_hits = set(
hit for hit in exclusive_hits
if self.single_hit_dist(hit) <= good_hit_threshold)
good_mutual_hits = set()
for i, j in (mutual_hits - exclusive_hits):
if self.hit_dists([(i, j)])[0] > good_hit_threshold:
continue
third_wheels = [
tup for tup in non_mutual_hits if i == tup[0] or j == tup[1]
]
if min(self.hit_dists(third_wheels)) > second_neighbor_thresh:
good_mutual_hits.add((i, j))
bad_mutual_hits = mutual_hits - exclusive_hits - good_mutual_hits
# --------------------------------------------------------------------------------
# Test that the four groups form a partition of all hits and finalize
# --------------------------------------------------------------------------------
assert (non_mutual_hits | bad_mutual_hits | good_mutual_hits
| exclusive_hits
== cluster_to_aligned_indexes | aligned_to_cluster_indexs_rev
and len(non_mutual_hits) + len(bad_mutual_hits) +
len(good_mutual_hits) + len(exclusive_hits)
== len(cluster_to_aligned_indexes
| aligned_to_cluster_indexs_rev))
self.non_mutual_hits = non_mutual_hits
self.mutual_hits = mutual_hits
self.bad_mutual_hits = bad_mutual_hits
self.good_mutual_hits = good_mutual_hits
self.exclusive_hits = exclusive_hits
if consider_tiles != 'all':
log.debug('Non-mutual hits: %s' % len(non_mutual_hits))
log.debug('Mutual hits: %s' % len(mutual_hits))
log.debug('Bad mutual hits: %s' % len(bad_mutual_hits))
log.debug('Good mutual hits: %s' % len(good_mutual_hits))
log.debug('Exclusive hits: %s' % len(exclusive_hits))
def least_squares_mapping(self, pct_thresh=0.9, min_hits=50):
"""least_squares_mapping(self, hit_type='exclusive')
"Input": set of tuples of (cluster_index, in_frame_idx) mappings.
"Output": scaling lambda, rotation theta, x_offset, y_offset, and aligned_rcs
We here solve the matrix least squares equation Ax = b, where
[ x0r -y0r 1 0 ]
[ y0r x0r 0 1 ]
A = [ x1r -y1r 1 0 ]
[ y1r x1r 0 1 ]
. . .
[ xnr -ynr 1 0 ]
[ ynr xnr 0 1 ]
and
b = [ x0s y0s x1s y1s . . . xns yns ]^T
The r and s subscripts indicate rcs and cluster coords.
The interpretation of x is then given by
x = [ alpha beta x_offset y_offset ]^T
where
alpha = lambda cos(theta), and
beta = lambda sin(theta)
This system of equations is then finally solved for lambda and theta.
"""
def get_hits(hit_type):
if isinstance(hit_type, str):
hit_type = [hit_type]
hits = []
for ht in hit_type:
hits.extend(getattr(self, ht + '_hits'))
return hits
found_good_mapping = False
for tile in self.hitting_tiles:
self.find_hits(consider_tiles=tile)
# Reminder: All indices are in the order (cluster_index, in_frame_idx)
raw_hits = get_hits(('exclusive', 'good_mutual'))
hits = self.remove_longest_hits(raw_hits, pct_thresh)
if len(hits) < min_hits:
continue
else:
found_good_mapping = True
A = np.zeros((2 * len(hits), 4))
b = np.zeros((2 * len(hits), ))
for i, (cluster_index, in_frame_idx) in enumerate(hits):
tile_key, (xir, yir) = self.rcs_in_frame[in_frame_idx]
A[2 * i, :] = [xir, -yir, 1, 0]
A[2 * i + 1, :] = [yir, xir, 0, 1]
xis, yis = self.clusters.point_rcs[cluster_index]
b[2 * i] = xis
b[2 * i + 1] = yis
alpha, beta, x_offset, y_offset = np.linalg.lstsq(A, b)[0]
offset = np.array([x_offset, y_offset])
theta = np.arctan2(beta, alpha)
lbda = alpha / np.cos(theta)
tile.set_aligned_rcs_given_transform(lbda, theta, offset)
tile.set_correlation(self.image_data.image)
if hasattr(self, 'control_corr'):
tile.set_snr_with_control_corr(self.control_corr)
return found_good_mapping
def rough_align(self,
possible_tile_keys,
rotation_est,
fq_w_est=927,
cluster_size=0.25,
snr_thresh=1.2):
self.fq_w = fq_w_est
self.set_fastq_tile_mappings()
self.set_all_fastq_image_data()
self.rotate_all_fastq_data(rotation_est)
start_time = time.time()
self.find_hitting_tiles(possible_tile_keys, cluster_size, snr_thresh)
log.debug('Rough alignment time: %.3f seconds' %
(time.time() - start_time))
def precision_align_only(self, min_hits):
start_time = time.time()
if not self.hitting_tiles:
raise RuntimeError('Alignment not found')
found_good_mapping = self.least_squares_mapping(min_hits=min_hits)
if not found_good_mapping:
raise ValueError("Could not precision align!")
log.debug('Precision alignment time: %.3f seconds' %
(time.time() - start_time))
start_time = time.time()
self.find_hits()
log.debug('Hit finding time: %.3f seconds' %
(time.time() - start_time))
@property
def alignment_stats(self):
hits = {
'exclusive': len(self.exclusive_hits),
'good_mutual': len(self.good_mutual_hits),
'bad_mutual': len(self.bad_mutual_hits),
'non_mutual': len(self.non_mutual_hits)
}
offsets = [
tuple(map(float, tile.offset)) for tile in self.hitting_tiles
]
return stats.AlignmentStats().from_data(
[str(tile.key) for tile in self.hitting_tiles],
[float(tile.scale) for tile in self.hitting_tiles],
[float(tile.width) for tile in self.hitting_tiles],
[float(tile.rotation_degrees)
for tile in self.hitting_tiles], offsets, hits)
@property
def read_names_rcs(self):
im_shape = self.image_data.image.shape
for tile in self.hitting_tiles:
if not hasattr(tile, 'rotation'):
# hack because I don't understand why tiles aren't getting rotations
# not having rotations implies they aren't getting aligned at all, which is very bad
continue
for read_name, pt in izip(tile.read_names, tile.aligned_rcs):
if 0 <= pt[0] < im_shape[0] and 0 <= pt[1] < im_shape[1]:
yield '%s\t%f\t%f\n' % (read_name, pt[0], pt[1])
def set_image_data(self, image, um_per_pixel):
self.image_data = ImageData(image.index, um_per_pixel, image)
if(verbose): print('\n\n##########Limpando Inúteis!\n#############################################')
for filename in os.listdir(path_final_biomemaps):
filename = os.path.splitext(filename)[0]
filename = filename.split('_')
ImageData.limpainutil(path_final_biomemaps+'biomemap_'+filename[1]+'.tif',path_final_heightmaps+'heightmap_'+filename[1]+'.tif')
'''
if(verbose): print('\n\n##########Gerando Descriptor!\n#############################################')
imagedata_list = []
descriptor = open(path_final+"imagedescriptor.txt","wb")
descriptor.truncate(0)
for filename in os.listdir(path_final_biomemaps):
filename = os.path.splitext(filename)[0]
filename = filename.split('_')
imagedata = ImageData(path_final_biomemaps+'biomemap_'+filename[1]+'.tif',path_final_heightmaps+'heightmap_'+filename[1]+'.tif')
imagedata_list.append(imagedata)
#if(verbose): print(imagedata.biomedata)
if(verbose): print('\n\n##########Salvando Descriptor!\n#############################################')
pickle.dump(imagedata_list,descriptor)
descriptor.close()
if(verbose): print('foram salvos '+str(len(imagedata_list))+' dados!')
if(verbose): print('\n\n##########Fim de Execução!\n#############################################')
def pick_n_images(self, n):
set_size = self.redis.scard(self.set_name)
random_members = self.redis.srandmember(self.set_name, min(n, set_size))
return [ImageData(None, None).deserialize(x) for x in random_members]