def UndirectedEdgeKey(image_index_a, image_index_b):
CHECK_GT(image_index_a, 0)
CHECK_GT(image_index_b, 0)
# ensure a is smaller than b
if image_index_a > image_index_b:
image_index_a, image_index_b = image_index_b, image_index_a
CHECK_GE(image_index_b, image_index_a)
key_pair = py_base.Uint64ToKey(image_index_a) + py_base.Uint64ToKey(
image_index_b)
return key_pair
def Run(self):
reader = py_pert.StringTableShardSetReader()
reader.Open(self.GetInput('images').GetUri())
image_ids = []
for i, (k, v) in enumerate(reader):
image_ids.append(ParseUint64Key(k))
LOG(INFO, 'creating match groups')
match_groups = [] # a list of tuples (primary_id, secondary id list)
widgets = [Percentage(), ' ', Bar(), ' ', ETA()]
pbar = ProgressBar(widgets=widgets, maxval=len(image_ids)).start()
for i in range(len(image_ids)):
primary_id = image_ids[i]
secondary_ids = list(image_ids)
secondary_ids.remove(primary_id)
for secondary_id_chunk in chunks(secondary_ids,
self.max_batch_size):
match_groups.append((primary_id, secondary_id_chunk))
pbar.update(i)
# write out the match plan (must be later sorted by key for future join stage)
writer = py_pert.StringTableWriter()
options = py_pert.WriterOptions()
options.SetUnsorted()
LOG(INFO, 'writing match groups')
CHECK(
writer.Open(
self.GetOutput('unsorted_match_batches').GetUri(), 1, options))
metadata = iw_pb2.MatchBatchMetadata()
pbar = ProgressBar(widgets=widgets, maxval=len(match_groups)).start()
for batch_id, (batch_primary_image,
batch_image_ids) in enumerate(match_groups):
if len(batch_image_ids) == 0:
continue
batch_name = py_base.Uint64ToKey(batch_id)
metadata = iw_pb2.MatchBatchMetadata()
metadata.image_id = batch_primary_image
metadata.batch_name = batch_name
metadata.is_primary = True
writer.Add(py_base.Uint64ToKey(metadata.image_id),
metadata.SerializeToString())
for image_id in batch_image_ids:
metadata.image_id = image_id
metadata.batch_name = batch_name
metadata.is_primary = False
writer.Add(py_base.Uint64ToKey(metadata.image_id),
metadata.SerializeToString())
pbar.update(batch_id)
return
def GetImage(self, image_id):
ok, value = self.image_reader.Find(py_base.Uint64ToKey(image_id))
if not ok:
return None
jpeg = iw_pb2.JpegImage()
jpeg.ParseFromString(value)
CHECK(jpeg.IsInitialized())
return jpeg
def __ImportImageArchive(self, extract_dir, dataset_root):
image_filenames = []
print 'Searching for image files...'
for root, dirnames, filenames in os.walk(extract_dir):
for filename in filenames:
if fnmatch.fnmatch(filename, '*.jpg') or fnmatch.fnmatch(
filename, '*.jpeg'):
image_filenames.append(os.path.join(root, filename))
pert_uri = 'local://%s/photoid_to_image.pert' % (dataset_root)
if len(image_filenames) <= 2:
return False
fingerprinted_path = '%s/fingerprinted/' % (extract_dir)
os.mkdir(fingerprinted_path)
print 'Fingerprinting image files...'
progress = util.MakeProgressBar(len(image_filenames))
# rename all files according to fingerprint
for i, filename in enumerate(image_filenames):
data = open(filename).read()
fp = py_base.FingerprintString(data)
dst = '%s/%064d.jpg' % (
fingerprinted_path, fp
) # ensure lexical sort = numeric sort = key sort
os.rename(filename, dst)
progress.update(i)
filenames = glob.glob('%s/*.jpg' % fingerprinted_path)
filenames.sort()
output_uri = 'local://%s/photoid_to_image.pert' % (dataset_root)
# write to pert in sorted order
print 'Generating image PERT file...'
writer = py_pert.StringTableWriter()
CHECK(writer.Open(output_uri, 1))
progress = util.MakeProgressBar(len(filenames))
for i, filename in enumerate(filenames):
data = open(filename).read()
key = py_base.Uint64ToKey(py_base.FingerprintString(data))
try:
im = Image.open(StringIO.StringIO(data))
except IOError as e:
LOG(INFO, 'Error opening %s - %s' % (filename, e))
continue
width, height = im.size
jpeg = iw_pb2.JpegImage()
jpeg.data = data
jpeg.width = width
jpeg.height = height
CHECK(jpeg.IsInitialized())
writer.Add(key, jpeg.SerializeToString())
progress.update(i)
writer.Close()
return True
def BatchGetImages(self, image_ids):
""" Does efficient batch lookup of images returning a dict mapping image_id to raw jpeg data. """
id_to_jpeg = {}
image_ids.sort()
for image_id in image_ids:
ok, value = self.image_reader.Find(py_base.Uint64ToKey(image_id))
CHECK(ok)
jpeg = iw_pb2.JpegImage()
jpeg.ParseFromString(value)
CHECK(jpeg.IsInitialized())
id_to_jpeg[image_id] = jpeg
return id_to_jpeg
def BatchGetImagesAsDataUri(self, image_ids):
""" Does efficient batch lookup of images returning a dict mapping image_id to raw jpeg data. """
id_to_datauri = {}
image_ids.sort()
self.image_reader.SeekToStart()
for image_id in image_ids:
ok, value = self.image_reader.Find(py_base.Uint64ToKey(image_id))
CHECK(ok, 'failed to find image_id: %d' % (image_id))
jpeg = iw_pb2.JpegImage()
jpeg.ParseFromString(value)
CHECK(jpeg.IsInitialized())
id_to_datauri[image_id] = JpegToDataUrl(jpeg.data)
return id_to_datauri
def main():
base_uri = 'local://home/ubuntu/Desktop/datasets/tide_v12/'
tide_uri = '%s/objectid_to_object.pert' % (base_uri)
dataset = tide.TideDataset(tide_uri)
print tide
pos_imageids = []
imageid_to_objectname = {}
for id, obj in dataset.objectid_to_object.iteritems():
print obj.name
pos_imageids.extend(obj.pos_image_ids)
for image_id in obj.pos_image_ids:
imageid_to_objectname[image_id] = obj.name
# sort for efficient access to pert
pos_imageids.sort()
images_pert_uri = '%s/photoid_to_image.pert' % (base_uri)
reader = py_pert.StringTableReader()
CHECK(reader.Open(images_pert_uri))
for image_id in pos_imageids:
ok, data = reader.Find(py_base.Uint64ToKey(image_id))
CHECK(ok)
jpeg_image = iw_pb2.JpegImage()
jpeg_image.ParseFromString(data)
objectname = imageid_to_objectname[image_id]
dirname = './extracted/%s' % (objectname)
filename = '%s/%d.jpg' % (dirname, image_id)
if not os.path.exists(dirname):
os.makedirs(dirname)
f = open(filename, 'wb')
f.write(jpeg_image.data)
return
def Uint64ToKey(i):
return py_base.Uint64ToKey(i)
def GenerateTestData():
for i in range(1000):
yield (py_base.Uint64ToKey(i), py_base.Uint64ToKey(i))
def RenderSvg(self, image_a_id, image_b_id):
svg = None
# look up match for this image pair
if image_a_id > image_b_id:
image_a_id, image_b_id = (image_b_id, image_a_id)
CHECK_LT(image_a_id, image_b_id)
match_result = iw_pb2.GeometricMatchResult()
match_key = py_base.Uint64ToKey(image_a_id) + py_base.Uint64ToKey(
image_b_id)
ok, value = self.match_reader.Find(match_key)
CHECK(ok, 'cant find key: %d %d' % (image_a_id, image_b_id))
match_result.ParseFromString(value)
# get image data
id_to_jpeg = self._BatchGetImages([image_a_id, image_b_id])
image_a = id_to_jpeg[image_a_id]
image_b = id_to_jpeg[image_b_id]
# render svg
width = image_a.width + image_b.width
height = max(image_a.height, image_b.height)
svg = """<?xml version="1.0" standalone="no"?><!DOCTYPE svg PUBLIC "-//W3C//DTD SVG 1.1//EN" "http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd">"""
svg += """<style> line { opacity: 0.25;} line:hover { opacity: 1.0;} </style>"""
svg += '<svg width="%d" height="%d" version="1.1" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink">' % (
width, height)
svg += '<image x="%dpx" y="%dpx" width="%dpx" height="%dpx" xlink:href="%s"> </image> \n' % (
0, 0, image_a.width, image_a.height, JpegToDataUrl(image_a.data))
svg += '<image x="%dpx" y="%dpx" width="%dpx" height="%dpx" xlink:href="%s"> </image> \n' % (
image_a.width, 0, image_b.width, image_b.height,
JpegToDataUrl(image_b.data))
colors = [
'yellow',
'red',
'blue',
]
match_info = []
for match_index, match in enumerate(match_result.matches):
match_info.append(match.nfa)
color = colors[match_index % len(colors)]
for correspondence_index, c in enumerate(match.correspondences):
a_x = c.a.pos.x
a_y = c.a.pos.y
b_x = c.b.pos.x + image_a.width
b_y = c.b.pos.y
#svg += '<line x1="%d" y1="%d" x2="%d" y2="%d" style="stroke:%s;stroke-width:2"/>\n' % (x1,y1,x2,y2, color)
left_pt_id = "lp%d" % (correspondence_index)
right_pt_id = "rp%d" % (correspondence_index)
svg += "<circle id=\"%s\" cx=\"%f\" cy=\"%f\" r=\"3\" stroke=\"black\" stroke-width=\"0\" fill=\"%s\"/>\n" % (
left_pt_id, a_x, a_y, color)
svg += "<circle id=\"%s\" cx=\"%f\" cy=\"%f\" r=\"3\" stroke=\"black\" stroke-width=\"0\" fill=\"%s\"/>\n" % (
right_pt_id, b_x, b_y, color)
svg += "<circle id=\"%s_support\" cx=\"%f\" cy=\"%f\" r=\"%f\" stroke-width=\"5\" fill=\"none\" opacity=\"0.5\" stroke=\"%s\" >\n" % (
left_pt_id, a_x, a_y, c.a.radius, color)
svg += "<set attributeName=\"opacity\" from=\"0.5\" to=\"1.0\" begin=\"%s.mouseover\" end=\"%s.mouseout\"/>" % (
left_pt_id, left_pt_id)
svg += "<set attributeName=\"opacity\" from=\"0.5\" to=\"1.0\" begin=\"%s.mouseover\" end=\"%s.mouseout\"/>" % (
right_pt_id, right_pt_id)
svg += "</circle>"
svg += "<circle id=\"%s_support\" cx=\"%f\" cy=\"%f\" r=\"%f\" stroke-width=\"5\" fill=\"none\" opacity=\"0.5\" stroke=\"%s\" >\n" % (
right_pt_id, b_x, b_y, c.b.radius, color)
svg += "<set attributeName=\"opacity\" from=\"0.5\" to=\"1.0\" begin=\"%s.mouseover\" end=\"%s.mouseout\"/>" % (
left_pt_id, left_pt_id)
svg += "<set attributeName=\"opacity\" from=\"0.5\" to=\"1.0\" begin=\"%s.mouseover\" end=\"%s.mouseout\"/>" % (
right_pt_id, right_pt_id)
svg += "</circle>"
svg += "<line x1=\"%f\" y1=\"%f\" x2=\"%f\" y2=\"%f\" style=\"stroke:rgb(255,0,0);stroke-width:2\" visibility=\"hidden\">" % (
a_x, a_y, b_x, b_y)
svg += "<set attributeName=\"visibility\" from=\"hidden\" to=\"visible\" begin=\"%s.mouseover\" end=\"%s.mouseout\"/>" % (
left_pt_id, left_pt_id)
svg += "<set attributeName=\"visibility\" from=\"hidden\" to=\"visible\" begin=\"%s.mouseover\" end=\"%s.mouseout\"/>" % (
right_pt_id, right_pt_id)
svg += "</line>"
svg += '</svg>'
for info in match_info:
svg += '<pre>nfa: %s</pre>' % (info)
return svg