def validation_check(verbose=0):
# Check on 100 images picked at random from validation set
val_captions=list()
for i in range(100):
idx=np.random.randint(0,1093)
c=generate_caption(image_dir+filenames_val[idx])
val_captions.append(copy(c))
if verbose:
print_progress(i+1,100)
return val_captions
def create_word_LUT():
word_lookup = list()
for i in range(len(bow)):
# for i in range(10):
word = bow[i]['word']
indices = images_with_word(captions_train, word)
entry = {'word': word, 'indices': indices}
word_lookup.append(copy(entry))
print_progress(i + 1, len(bow))
return word_lookup
def bulk_generation_KNN():
num_test_images = np.shape(filenames_test)[0]
generated_captions = list()
for i in range(num_test_images):
image_name = filenames_test[i]
if i == 884:
image_name = filenames_test[883]
caption = KNN_score(image_name)
generated_captions.append(caption)
print_progress(i, num_test_images)
return generated_captions
def generate_captions_testset(alpha):
print("Generating captions for alpha =", alpha)
generated_captions = list()
for i in range(len(transfer_values_test)):
transfer_value = transfer_values_test[i]
generated_caption = generate_caption_reference(transfer_value,
alpha=alpha)
generated_captions.append(generated_caption)
print_progress(i + 1, len(transfer_values_test))
filename = 'generated_captions_reference_alpha_' + str(alpha) + '.json'
with open(filename, 'w') as f:
json.dump(generated_captions, f)
s4 = sentence_bleu(references[i],
candidate_tokenized,
weights=[0.25, 0.25, 0.25, 0.25],
smoothing_function=chencherry.method1)
# First approach: the best sentence is the one with the best
# BLEU sum
bleuSum = s1 + s2 + s3 + s4
bleuSums.append(bleuSum)
bestIdx = np.argmax(bleuSums)
# print(bestCaption,bleuSums)
bestCaption = beamCaptions[i][bestIdx]['sentence']
bestCaptions.append(bestCaption[1:])
idxList.append(bestIdx)
# plot progress
print_progress(i + 1, num_samples)
with open(out_dir + 'absolute_best.json', 'w') as outFile:
json.dump(bestCaptions, outFile)
# SECON METHOD
# the best caption is the one with the best consensus score
# VGG16 consensus
print('\nChoosing best caption based on VGG16 consensus')
image_dir = 'UAV/images/'
VGG16_idxList = list()
VGG16_bestCaptions = list()
transfer_values_train = np.load(
def parse_links(names, src, menu_links, input_path, output_path,
layer_level, recursion_depth):
"""
Parses links and organizes folders recursively
- Finds suitable links in html file corresponding to input path
- Changes the links by giving them unique names
- Creates a new directory with that unique name in the output_path
- Generates the modified html file in the output_path
"""
print_progress(layer_level)
with open(input_path) as f:
soup = BeautifulSoup(f.read())
html_path = os.path.join(output_path, "index.html")
parse_css_links(soup, layer_level)
parse_images(soup, layer_level)
save_html(soup, html_path)
bread_tag = soup.find("div", attrs={"class": "breadcum"})
bread_count = 0
if bread_tag:
if bread_tag.find_all('a'):
for link_tag in reversed(bread_tag.find_all('a')):
link_tag['href'] = bread_count*"../"
if layer_level is 3:
link_tag['href'] = bread_count*"../" + "../"
bread_count = bread_count + 1
for link_tag in soup.find_all('a'):
parse_menu_links(layer_level, link_tag, menu_links)
link = link_tag.get('href')
if link is None:
continue
elink = link_tag.get('href')
if link.startswith("http://"):
if (link in downloadablelinks):
link = link[6:]
link_tag['href'] = layer_level*"../" + "../external" + link
if layer_level is 3:
link_tag['href'] = 4*"../" + "../external" + link
elif (link in activeallexternallinks):
link_tag['href'] = 'javascript:var c=confirm("Do You wish to access internet?");if(c==true){window.location="'+elink+'";}'
else:
link_tag['href'] = 'javascript:alert("Link is dead");'
if link_is_appropriate(link, layer_level):
name = link_tag.string.strip(" \t\n\r").partition(' ')[0].lower()
name = get_unique_name(name, names)
if layer_level is 2:
name = name + "/theory"
link_tag['href'] = name
new_output_path = os.path.join(output_path, name)
make_new_directory(new_output_path)
if recursion_depth > 0:
parse_links(names, src, menu_links,
input_path=os.path.join(src, link),
output_path=new_output_path,
layer_level=layer_level + 1,
recursion_depth=recursion_depth - 1)
if (layer_level is 1 and link.count('&') is 3):
name2 = link_tag.string.strip(" \t\n\r").partition(' ')[0].lower()
link_tag['href'] = name + "/" + name2 + "/theory"
parse_last_layer(link_tag, names, src, menu_links, input_path,
output_path, layer_level, recursion_depth)
save_html(soup, html_path)
def parse_links(names, src, menu_links, input_path, output_path, layer_level,
recursion_depth):
"""
Parses links and organizes folders recursively
- Finds suitable links in html file corresponding to input path
- Changes the links by giving them unique names
- Creates a new directory with that unique name in the output_path
- Generates the modified html file in the output_path
"""
print_progress(layer_level)
with open(input_path) as f:
soup = BeautifulSoup(f.read())
html_path = os.path.join(output_path, "index.html")
parse_css_links(soup, layer_level)
parse_images(soup, layer_level)
save_html(soup, html_path)
bread_tag = soup.find("div", attrs={"class": "breadcum"})
bread_count = 0
if bread_tag:
if bread_tag.find_all('a'):
for link_tag in reversed(bread_tag.find_all('a')):
link_tag['href'] = bread_count * "../"
if layer_level is 3:
link_tag['href'] = bread_count * "../" + "../"
bread_count = bread_count + 1
for link_tag in soup.find_all('a'):
parse_menu_links(layer_level, link_tag, menu_links)
link = link_tag.get('href')
if link is None:
continue
elink = link_tag.get('href')
if link.startswith("http://"):
if (link in downloadablelinks):
link = link[6:]
link_tag['href'] = layer_level * "../" + "../external" + link
if layer_level is 3:
link_tag['href'] = 4 * "../" + "../external" + link
elif (link in activeallexternallinks):
link_tag[
'href'] = 'javascript:var c=confirm("Do You wish to access internet?");if(c==true){window.location="' + elink + '";}'
else:
link_tag['href'] = 'javascript:alert("Link is dead");'
if link_is_appropriate(link, layer_level):
name = link_tag.string.strip(" \t\n\r").partition(' ')[0].lower()
name = get_unique_name(name, names)
if layer_level is 2:
name = name + "/theory"
link_tag['href'] = name
new_output_path = os.path.join(output_path, name)
make_new_directory(new_output_path)
if recursion_depth > 0:
parse_links(names,
src,
menu_links,
input_path=os.path.join(src, link),
output_path=new_output_path,
layer_level=layer_level + 1,
recursion_depth=recursion_depth - 1)
if (layer_level is 1 and link.count('&') is 3):
name2 = link_tag.string.strip(" \t\n\r").partition(' ')[0].lower()
link_tag['href'] = name + "/" + name2 + "/theory"
parse_last_layer(link_tag, names, src, menu_links, input_path,
output_path, layer_level, recursion_depth)
save_html(soup, html_path)
