def calculate_and_write_edge_weigthings_for_synsets(synset_filenames_dict, file_name):
max_co_occurrence = calculate_max_co_occurrence(synset_filenames_dict)
edge_weigthings_for_synsets = dict()
how_many_added = 0
how_many_done = 0
how_many_to_do = len(synset_filenames_dict.keys()) * (len(synset_filenames_dict.keys())-1)
write_edge_weightings_to_file(dict(), file_name)
for synset1, filenames1 in synset_filenames_dict.iteritems():
for synset2, filenames2 in synset_filenames_dict.iteritems():
if synset1 < synset2:
how_many_done += 1
#if (synset1.name, synset2.name) not in similarity_histogram:
similarity = wn.synset(synset1).lch_similarity(wn.synset(synset2))
co_occurence = len(set(synset_filenames_dict[synset1]).intersection(set(synset_filenames_dict[synset2])))
normalized_co_occurrence = co_occurence/max_co_occurrence
if similarity < 2.0:
similarity = 0
if normalized_co_occurrence < 0.4:
normalized_co_occurrence = 0
edge_weighting = similarity + 4*normalized_co_occurrence
if edge_weighting != 0:
edge_weigthings_for_synsets[(synset1, synset2)] = edge_weighting
how_many_added += 1
if how_many_added > 1000:
print_status("Done with " + str(how_many_done) + " von " + str(how_many_to_do) + "\n")
write_edge_weightings_to_file(edge_weigthings_for_synsets, file_name, append_to_file=True)
edge_weigthings_for_synsets = dict()
how_many_added = 0
write_edge_weightings_to_file(edge_weigthings_for_synsets, file_name, append_to_file=True)
def keyword_clustering_via_mcl(synset_filenames_dict):
config = ConfigParser.SafeConfigParser()
config.read('../config.cfg')
edge_weightings_filename = config.get('Filenames for Pickles', 'edge_weightings_filename')
calculate_and_write_edge_weigthings_for_synsets(synset_filenames_dict, edge_weightings_filename)
print_status("Done calculation of edge weightings and writing to file.\n")
print_status("Start MCL Clustering...")
mcl_clustering(edge_weightings_filename)
print "Done."
def write_edge_weightings_to_file(edge_weigthings_for_synsets, file_name, append_to_file=False):
print_status("Writing edge weightings to mcl readable file.")
output_file = None
if append_to_file:
output_file = open(file_name, 'a')
else:
output_file = open(file_name, 'w')
for (synset1, synset2), edge_weighting in edge_weigthings_for_synsets.iteritems():
output_file.write(str(synset1) + ' ' + str(synset2) + ' ' + str(edge_weighting) + '\n')
output_file.close()
print "Done."
def calculate_average_distance(parsed_result_tree, image_id_tuples, id_type, verbose=False):
distances = []
for image_id_1, image_id_2 in image_id_tuples:
if verbose: print_status("Checking for ids %s and %s (%s)... " % (image_id_1, image_id_2, id_type))
distance = find_closest_match_to_nodes(parsed_result_tree, image_id_1, image_id_2)
if distance != float('inf'):
distances.append(distance)
if verbose: sys.stdout.write("distance is: %s\n" % distance)
else:
if verbose: sys.stdout.write("one image could not be found!\n")
return float(sum(distances)) / len(distances)
def main(args):
print_status("Checking images against testset:\n")
print_status("Loading visual_features from file... ")
visual_features = general_helpers.load_visual_features()
print "Done."
print_status("Loading cluster_for_synsets from mcl_clusters file... ")
cluster_for_synsets = general_helpers.load_cluster_for_synsets()
print "Done."
print_status("Loading keywords_for_pictures from file... ")
keywords_for_pictures = general_helpers.load_keywords_for_pictures()
print "Done."
print_status("Loading cluster_representatives from file... ")
cluster_representatives = general_helpers.load_cluster_representatives(how_many_per_cluster=6)
print "Done loading preprocessed data."
print_status("Checking images against testset:\n")
print_status("Retrieving clusters... \n")
# image_tree = get_searchtrees_with_filenames("food", use_meronyms=False, minimal_node_size=1)
image_tree = pipeline.get_clusters(
"food",
use_meronyms=False,
visual_clustering_threshold=10000,
mcl_clustering_threshold=15,
minimal_mcl_cluster_size=10,
minimal_node_size=15,
visual_features=visual_features,
cluster_for_synsets=cluster_for_synsets,
keywords_for_pictures=keywords_for_pictures,
cluster_representatives=cluster_representatives,
)
sys.stdout.write("Collecting images from tree... \n")
result_ids = recursively_collect_images(image_tree)
sys.stdout.write("Loading testset from database... \n")
testset_positive_ids, testset_negative_ids = retrieveTestsetResults(args.database_file)
sys.stdout.write("Comparing result images to testset... \n")
result_size = len(result_ids)
testset_positive_size = len(testset_positive_ids)
testset_negative_size = len(testset_negative_ids)
true_positives = 0
false_positives = 0
for result_id in result_ids:
if result_id in testset_positive_ids:
true_positives += 1
testset_positive_ids.remove(result_id)
if result_id in testset_negative_ids:
false_positives += 1
testset_negative_ids.remove(result_id)
false_negatives = len(testset_positive_ids)
precision = float(true_positives) / (true_positives + false_positives)
recall = float(true_positives) / (true_positives + false_negatives)
sys.stdout.write("Done:\n\n")
sys.stdout.write("Testset size: %d\n\n" % (testset_positive_size + testset_negative_size))
sys.stdout.write("Result size: %d\n" % result_size)
sys.stdout.write("Real positives: %d\n\n" % testset_positive_size)
sys.stdout.write("True Positives: %d\n" % true_positives)
sys.stdout.write("True Negatives: ???\n")
sys.stdout.write("False Positives: %d\n" % false_positives)
sys.stdout.write("False Negatives: %d\n\n" % false_negatives)
sys.stdout.write("Precision: %f (tp / (tp + fp))\n" % precision)
sys.stdout.write("Recall: %f (tp / (tp + fn))\n" % recall)
sys.stdout.write(
"F-Measure: %f (2 * (p * r / (p + r)))\n"
% (2 * (float(precision) * float(recall)) / (precision + recall))
)
def main(args):
# Loading preprocessed features on startup
print_status("Loading visual_features from file... ")
visual_features = general_helpers.load_visual_features()
print "Done."
print_status("Loading cluster_for_synsets from mcl_clusters file... ")
cluster_for_synsets = general_helpers.load_cluster_for_synsets()
print "Done."
print_status("Loading keywords_for_pictures from file... ")
keywords_for_pictures = general_helpers.load_keywords_for_pictures()
print "Done."
print_status("Loading cluster_representatives from file... ")
cluster_representatives = general_helpers.load_cluster_representatives(how_many_per_cluster=6)
print "Done loading preprocessed data."
print_status("Checking images against testset:\n")
print_status("Retrieving clusters... \n")
pipeline_result = pipeline.get_clusters("food", use_meronyms=False,
visual_clustering_threshold=100000,
mcl_clustering_threshold=10,
minimal_mcl_cluster_size=1,
minimal_node_size=10,
visual_features=visual_features,
cluster_for_synsets=cluster_for_synsets,
keywords_for_pictures=keywords_for_pictures,
cluster_representatives=cluster_representatives)
# pipeline_result = pickle.load(open('image_tree.pickle', 'r'))
print_status("Parsing result tree to easier accessible format...")
parsed_result_tree = parse_result_tree(pipeline_result)
print_status("Loading testset from database... \n")
same_object_ids, same_object_same_context_ids, not_similar_ids = retrieveTestsetResults(args.database_file)
print_status("Comparing result images to testset... \n")
average_same_object_distance = calculate_average_distance(parsed_result_tree, same_object_ids, "same object", verbose=True)
average_same_context_distance = calculate_average_distance(parsed_result_tree, same_object_same_context_ids, "same context", verbose=True)
average_not_similar_distance = calculate_average_distance(parsed_result_tree, not_similar_ids, "not_similar", verbose=True)
print_status("Done!\n")
sys.stdout.write("Average distance for same object is %s with closeness %s \n" % (average_same_object_distance, float(1)/average_same_object_distance))
sys.stdout.write("Average distance for same context is %s with closeness %s \n" % (average_same_context_distance, float(1)/average_same_context_distance))
sys.stdout.write("Average distance for not similar is %s with closeness %s \n" % (average_not_similar_distance, float(1)/average_not_similar_distance))
sys.stdout.write("Distance %s \n" % (float(1)/average_same_object_distance - float(1)/average_not_similar_distance))
def main(args):
# Loading preprocessed features on startup
print_status("Loading visual_features from file... ")
visual_features = general_helpers.load_visual_features()
print "Done."
print_status("Loading cluster_for_synsets from mcl_clusters file... ")
cluster_for_synsets = general_helpers.load_cluster_for_synsets()
print "Done."
print_status("Loading keywords_for_pictures from file... ")
keywords_for_pictures = general_helpers.load_keywords_for_pictures()
print "Done."
print_status("Loading cluster_representatives from file... ")
cluster_representatives = general_helpers.load_cluster_representatives(how_many_per_cluster=6)
print "Done loading preprocessed data."
print_status("Checking images against testset:\n")
print_status("Retrieving clusters... \n")
pipeline_result = pipeline.get_clusters("food", use_meronyms=False,
visual_clustering_threshold=100000,
mcl_clustering_threshold=4,
minimal_mcl_cluster_size=6,
minimal_node_size=4,
visual_features=visual_features,
cluster_for_synsets=cluster_for_synsets,
keywords_for_pictures=keywords_for_pictures,
cluster_representatives=cluster_representatives)
# # Comment in to load preprocessed pipeline_result for dev mode
# pipeline_result = pickle.load(open('image_tree.pickle', 'r'))
annotated_food_dict = json.load(open(args.food_id_file, 'r'))
print_status("Flattening result tree... \n")
flattened_mcl_tree = flatten_result_tree(pipeline_result, annotated_food_dict, size_from_id=0, size_to_id=-1)
image_counter = len(flattened_mcl_tree.subclusters[0]['subcluster'])
print_status("Loading visual_features from file... \n")
visual_features = general_helpers.load_visual_features()
true_positives_total = []
false_negatives_total = []
true_negatives_total = []
false_positives_total = []
for i in range(0, 10):
print_status("Calculating visual clusters (%d x)... \n" % i)
visually_clustered_result = combined_clustering.cluster_visually(copy.deepcopy(flattened_mcl_tree),
visual_clustering_threshold=4,
visual_features=visual_features)
print_status("Convert visual clusters to simpler structure... \n")
visual_clusters = []
for visual_cluster in visually_clustered_result.subclusters[0]['subcluster']:
visual_clusters.append(set([image_tuple[0].split('\\')[-1].split('.')[0] for image_tuple in visual_cluster]))
print_status("Done clustering %d images into %d visual clusters. \n" % (image_counter, len(visual_clusters)))
# # Comment in to load preprocessed visual_clusters for dev mode
# visual_clusters = pickle.load(open('visual_clusters.pickle', 'r'))
print_status("Loading testset from database... \n")
visually_similar_tuples, visually_different_tuples = retrieveTestsetResults(args.database_file)
print_status("Comparing clusters to testset... \n")
true_negatives = 0
false_positives = 0
true_positives = 0
false_negatives = 0
print_status("Starting with visually similar tuples... \n")
for id_tuple in visually_similar_tuples:
if both_ids_are_found(id_tuple, visual_clusters):
if one_cluster_contains_both_ids(id_tuple, visual_clusters):
true_negatives += 1
else:
false_positives += 1
print_status("Now checking different image tuples... \n")
for id_tuple in visually_different_tuples:
if both_ids_are_found(id_tuple, visual_clusters):
if one_cluster_contains_both_ids(id_tuple, visual_clusters):
false_negatives += 1
else:
true_positives += 1
true_positives_total.append(true_positives)
false_negatives_total.append(false_negatives)
true_negatives_total.append(true_negatives)
false_positives_total.append(false_positives)
average_true_positives = float(sum(true_positives_total)) / len(true_positives_total)
average_false_negatives = float(sum(false_negatives_total)) / len(false_negatives_total)
average_true_negatives = float(sum(true_negatives_total)) / len(true_negatives_total)
average_false_positives = float(sum(false_positives_total)) / len(false_positives_total)
precision = float(average_true_positives) / (average_true_positives + average_false_positives)
recall = float(average_true_positives) / (average_true_positives + average_false_negatives)
print_status("Done!\n\n")
sys.stdout.write("Testset contains %5d visually similar image tuples \n" % len(visually_similar_tuples))
sys.stdout.write("And there are %5d visually different image tuples \n\n" % len(visually_different_tuples))
sys.stdout.write("Average true positives: %f \n" % average_true_positives)
sys.stdout.write("Average false negatives: %f \n" % average_false_negatives)
sys.stdout.write("Average true negatives: %f \n" % average_true_negatives)
sys.stdout.write("Average false positives: %f \n\n" % average_false_positives)
sys.stdout.write("Precision: %f (tp / (tp + fp))\n" % precision)
sys.stdout.write("Recall: %f (tp / (tp + fn))\n" % recall)
sys.stdout.write("F-Measure: %f (2 * (p * r / (p + r)))\n" % (2 * (float(precision) * float(recall)) / (precision + recall)))
import sys
from flask import Flask, url_for, Response
from flask import render_template
from flask_assets import Environment, Bundle
import string
# import own modules
from clustering.pipeline import get_clusters
from helpers.general_helpers import print_status, load_visual_features, load_cluster_for_synsets, load_keywords_for_pictures
from helpers.general_helpers import load_cluster_representatives
app = Flask(__name__)
assets = Environment(app)
# Loading preprocessed features on startup
print_status("Loading cluster_for_synsets from mcl_clusters file... ")
cluster_for_synsets = load_cluster_for_synsets()
print "Done."
print_status("Loading keywords_for_pictures from file... ")
keywords_for_pictures = load_keywords_for_pictures()
print "Done."
print_status("Loading cluster_representatives from file... ")
cluster_representatives = load_cluster_representatives(how_many_per_cluster=6)
print "Done.\n\n"
bufferedSearches = {}
print_status("Server is ready!\n\n")
@app.route("/")
def hello():
return render_template('index.html')
