default=150000)
optionParser.add_option('-t',
dest='num_iterations',
help="Number of k-means iterations",
type=int,
default=15)
(options,args) = optionParser.parse_args()
# -- Usage
if None in [options.feature_file, options.index_file, options.num_clusters]:
optionParser.print_help()
sys.exit(0)
# -- Load the feature file
log.info("Loading the feature file...")
f = FeatureList(options.feature_file)
f.process()
log.debug("Loaded %s rows" % len(f.dataset))
# -- Load the index
log.info("Loading the index file...")
flann = FLANN(log_level='info')
flann.load_index(options.index_file,f.dataset)
# -- Cluster all of the points
log.info("Clustering into %i clusters, over %i iterations" %
(options.num_clusters,options.num_iterations))
cluster_data = flann.kmeans(f.dataset,options.num_clusters,dtype=uint8,
max_iterations=options.num_iterations)
# -- Save the clusters to file
image_list = file(opts.image_list,'r')
image_filenames = [line.strip() for line in image_list.readlines()]
# -- Read in the number of features per file
feature_count = dict()
feature_size_list = file(opts.feature_size_list,'r')
for image in image_filenames:
feature_count[image] = int( feature_size_list.readline() )
log.info( "Loaded information for %i images" % len(feature_count) )
# -- Sanity check. The first image, has 380 features.
assert 380 == feature_count['00004fc7eb4bf00ba434c167890b99fa.jpg']
# -- Load the feature file
log.info("Loading features from file...")
f = FeatureList(opts.feature_file)
f.process()
log.debug("Loaded %s rows" % len(f.dataset))
# -- Load the clusters
log.info("Loading clusters from file")
cluster_file = file(opts.cluster_file,'r')
base = basename(opts.cluster_file)
clusters = imp.load_source(base,opts.cluster_file).clusters
# -- Load the clusters into FLANN as if they were the points
log.info("Building cluster index...")
flann = buildIndex(clusters)
# -- For each point in the feature list, find its 'nearest neighbor',
# i.e. which cluster it belongs to.
image_list = file(opts.image_list, 'r')
image_filenames = [line.strip() for line in image_list.readlines()]
# -- Read in the number of features per file
feature_count = dict()
feature_size_list = file(opts.feature_size_list, 'r')
for image in image_filenames:
feature_count[image] = int(feature_size_list.readline())
log.info("Loaded information for %i images" % len(feature_count))
# -- Sanity check. The first image, has 380 features.
assert 380 == feature_count['00004fc7eb4bf00ba434c167890b99fa.jpg']
# -- Load the feature file
log.info("Loading features from file...")
f = FeatureList(opts.feature_file)
f.process()
log.debug("Loaded %s rows" % len(f.dataset))
# -- Load the clusters
log.info("Loading clusters from file")
cluster_file = file(opts.cluster_file, 'r')
base = basename(opts.cluster_file)
clusters = imp.load_source(base, opts.cluster_file).clusters
# -- Load the clusters into FLANN as if they were the points
log.info("Building cluster index...")
flann = buildIndex(clusters)
# -- For each point in the feature list, find its 'nearest neighbor',
# i.e. which cluster it belongs to.
