def Comic():
df = None
classes = ['bicycle','bird','car','cat','dog','person']
sets = [('comic','train'),('comic','test')]
for base,image_set in sets:
path_b = '/media/gonthier/HDD/data/cross-domain-detection/datasets/comic/ImageSets/Main/%s.txt'%(image_set)
pd_b = pd.read_csv(path_b,sep=r"\s*",names=['name_img'],dtype=str)
for c in classes:
pd_b[c] = -1
print(pd_b.head(5))
for index, row in pd_b.iterrows():
i = row['name_img']
path_i = '/media/gonthier/HDD/data/cross-domain-detection/datasets/comic/Annotations/%s.xml'%(i)
read_file = voc_eval.parse_rec(path_i)
for element in read_file:
classe_elt = element['name']
for c in classes:
if classe_elt==c:
pd_b.loc[pd_b['name_img']==row['name_img'],c] = 1
pd_b['set'] = image_set
if df is None:
df = pd_b
else:
df = df.append(pd_b)
output_name = path_output + 'comic_all' + '.csv'
df.to_csv(output_name)
output_name = path_output + 'comic' + '.csv'
df.to_csv(output_name)
def WriteDifficultsBoxes():
"""
This function will mark as difficult all the tiny objects in the xml files
"""
size_min = 25*25 # 20*20 Au moins un truc de taille superieur a 17*17
path_b ='/media/gonthier/HDD/data/Wikidata_Paintings/WikiTenLabels/Main/test.txt'
path_to_im = '/media/gonthier/HDD/data/Wikidata_Paintings/WikiTenLabels/JPEGImages/'
pd_b = pd.read_csv(path_b,sep=r"\s*",names=['item'],dtype=str)
for index, row in pd_b.iterrows():
Erase = False
i = row['item']
path_i = path_to_im + i +'.jpg'
im = cv2.imread(path_i)
height = im.shape[0]
width = im.shape[1]
writer = Writer(path_i, width, height)
pathxml = '/media/gonthier/HDD/data/Wikidata_Paintings/WikiTenLabels/Annotations/%s.xml'%(i)
read_file = voc_eval.parse_rec(pathxml)
for element in read_file:
classe_elt = element['name']
bbox = element['bbox']
xmin = bbox[0]
ymin = bbox[1]
xmax = bbox[2]
ymax = bbox[3]
area = (xmax -xmin)*(ymax-ymin)
if area <= size_min:
# Marked as difficult
element['difficult']=1
Erase = True
writer.addObject(classe_elt, xmin, ymin, xmax, ymax, difficult=1)
else:
writer.addObject(classe_elt, xmin, ymin, xmax, ymax)
if Erase:
writer.save(annotation_path=pathxml)
print('Modified :',i)
return(0)
def Clipart():
df = None
classes = ["aeroplane", "bicycle", "bird", "boat", "bottle", "bus", "car", "cat", "chair", "cow", "diningtable", "dog", "horse", "motorbike", "person", "pottedplant", "sheep", "sofa", "train", "tvmonitor"]
sets = [('clipart','train'),('clipart','test')]
for base,image_set in sets:
path_b = '/media/gonthier/HDD/data/cross-domain-detection/datasets/clipart/ImageSets/Main/%s.txt'%(image_set)
pd_b = pd.read_csv(path_b,sep=r"\s*",names=['name_img'],dtype=str)
for c in classes:
pd_b[c] = -1
print(pd_b.head(5))
for index, row in pd_b.iterrows():
i = row['name_img']
path_i = '/media/gonthier/HDD/data/cross-domain-detection/datasets/clipart/Annotations/%s.xml'%(i)
read_file = voc_eval.parse_rec(path_i)
for element in read_file:
classe_elt = element['name']
for c in classes:
if classe_elt==c:
pd_b.loc[pd_b['name_img']==row['name_img'],c] = 1
pd_b['set'] = image_set
if df is None:
df = pd_b
else:
df = df.append(pd_b)
output_name = path_output + 'clipart_all' + '.csv'
print(df.iloc[[45,46,47]])
df.to_csv(output_name)
output_name = path_output + 'clipart' + '.csv'
# On remplace les 0 par des 1 ! les cas difficiles par des certitudes
for c in classes:
df.loc[df[c]==0,c] = 1
print(df.iloc[[45,46,47]])
df.to_csv(output_name)
df=pd.read_csv(output_name,dtype=str)
print(df.iloc[[45,46,47]])
def Watercolor():
df = None
classes = ["bicycle", "bird","car", "cat", "dog", "person"]
sets = [('watercolor','train'),('watercolor','test')]
for base,image_set in sets:
path_b = '/media/gonthier/HDD/data/cross-domain-detection/datasets/watercolor/ImageSets/Main/%s.txt'%(image_set)
pd_b = pd.read_csv(path_b,sep=r"\s*",names=['name_img'],dtype=str)
for c in classes:
pd_b[c] = -1
print(pd_b.head(5))
for index, row in pd_b.iterrows():
i = row['name_img']
path_i = '/media/gonthier/HDD/data/cross-domain-detection/datasets/watercolor/Annotations/%s.xml'%(i)
read_file = voc_eval.parse_rec(path_i)
for element in read_file:
classe_elt = element['name']
for c in classes:
if classe_elt==c:
pd_b.loc[pd_b['name_img']==row['name_img'],c] = 1
pd_b['set'] = image_set
if df is None:
df = pd_b
else:
df = df.append(pd_b)
output_name = path_output + 'watercolor_all' + '.csv'
print(df.iloc[[45,46,47]])
df.to_csv(output_name)
output_name = path_output + 'watercolor' + '.csv'
# On remplace les 0 par des 1 ! les cas difficiles par des certitudes
for c in classes:
df.loc[df[c]==0,c] = 1
print(df.iloc[[45,46,47]])
df.to_csv(output_name)
df=pd.read_csv(output_name,dtype=str)
print(df.iloc[[45,46,47]])
def StatsOnWikiTenLabels():
annotations_folder = '/media/gonthier/HDD/data/Wikidata_Paintings/WikiTenLabels/Annotations/'
path_data = '/media/gonthier/HDD/output_exp/ClassifPaintings/'
name_file = path_data + 'WikiTenLabels.csv'
classes_a_garder = ['angel','Child_Jesus','crucifixion_of_Jesus','Mary','nudity', 'ruins','Saint_Sebastien']
df = pd.read_csv(name_file,sep=',')
df_test = df[df['set']=='test']
df_train = df[df['set']=='train']
df_test[classes_a_garder] = df_test[classes_a_garder]
print("For test")
df_test['sum'] = df_test[classes_a_garder].sum(axis=1)
for i in range(len(classes_a_garder)):
print(i,len(np.where(np.array(df_test.as_matrix(['sum']).ravel(),dtype=int)==i)[0]))
df_train[classes_a_garder] = df_train[classes_a_garder]
print("For train")
df_train['sum'] = df_train[classes_a_garder].sum(axis=1)
for i in range(len(classes_a_garder)):
print(i,len(np.where(np.array(df_train.as_matrix(['sum']).ravel(),dtype=int)==i)[0]))
print('Statistiques sur le test set')
print(df_test.sum())
print()
print('Statistiques sur le train set')
print(df_train.sum())
# classes = ['angel', 'beard','capital','Child_Jesus', 'crucifixion_of_Jesus',
# 'Mary','nudity', 'ruins','Saint_Sebastien','turban']
# list_elt= os.listdir(annotations_folder)
# file_test = '/media/gonthier/HDD/data/Wikidata_Paintings/WikiTenLabels/ImageSets/Main/test.txt'
# file = open(file_test,"w")
# for elt in list_elt:
# elt_wt_jpg = elt.split('.')[0]
# str_w = elt_wt_jpg +' \n'
# file.write(str_w)
# file.close()
size_min = 25*25 # 15*15
path_b ='/media/gonthier/HDD/data/Wikidata_Paintings/WikiTenLabels/ImageSets/Main/test.txt'
pd_b = pd.read_csv(path_b,sep=r"\s*",names=['item'],dtype=str)
dict_elts_total = {}
dict_elts_sizemin = {}
for c in classes_a_garder:
pd_b[c] = 0
dict_elts_total[c] = 0
dict_elts_sizemin[c] = 0
without_class = 0
numberofIm = 0
for index, row in pd_b.iterrows():
numberofIm += 1
i = row['item']
path_i = '/media/gonthier/HDD/data/Wikidata_Paintings/WikiTenLabels/Annotations/%s.xml'%(i)
read_file = voc_eval.parse_rec(path_i)
with_class = False
for element in read_file:
classe_elt = element['name']
bbox = element['bbox']
xmin = bbox[0]
ymin = bbox[1]
xmax = bbox[2]
ymax = bbox[3]
area = (xmax -xmin)*(ymax-ymin)
# print(area)
for c in classes_a_garder:
if classe_elt==c: # We got an element from
with_class = True
pd_b.loc[pd_b['item']==row['item'],c] = 1
dict_elts_total[c] += 1
if area > size_min:
dict_elts_sizemin[c] += 1
if not(with_class):
without_class += 1
print('Statistiques au niveaux du nombre de classes avec les labels dans la partie annotee en detection')
print(pd_b.sum())
print('Nombre d instances des differentes classes')
num_obj = 0
for c in classes_a_garder:
print(c,' : ',dict_elts_total[c])
num_obj+=dict_elts_total[c]
print('Nombre d images totales',numberofIm)
print('Nombre d instances totales',num_obj)
print("Nombre d images sans classes",without_class)
print('Nombre d instances des differentes classes avec une taille superieur a :',size_min,'pixels',num_obj)
num_obj = 0
for c in classes_a_garder:
print(c,' : ',dict_elts_sizemin[c])
num_obj+=dict_elts_sizemin[c]
print('Nombre d instances de taille superieur a ',size_min,'pixels',num_obj)
def Stats_and_testFile():
annotations_folder = '/media/gonthier/HDD/data/Wikidata_Paintings/WikiTenLabels/Annotations/'
path_data = '/media/gonthier/HDD/output_exp/ClassifPaintings/'
name_file = path_data + 'WikiTenLabels.csv'
df = pd.read_csv(name_file,sep=',')
df_test = df[df['set']=='test']
print('Statistiques sur le test set')
print(df_test.sum())
classes = ['angel', 'beard','capital','Child_Jesus', 'crucifixion_of_Jesus',
'Mary','nudity', 'ruins','Saint_Sebastien','turban']
list_elt= os.listdir(annotations_folder)
write_test_file = False
if write_test_file:
file_test = '/media/gonthier/HDD/data/Wikidata_Paintings/WikiTenLabels/Main/ImageSets/test.txt'
file = open(file_test,"w")
for elt in list_elt:
elt_wt_jpg = elt.split('.')[0]
str_w = elt_wt_jpg +' \n'
file.write(str_w)
file.close()
size_min = 25*25 # 15*15
path_b ='/media/gonthier/HDD/data/Wikidata_Paintings/WikiTenLabels/ImageSets/Main/test.txt'
pd_b = pd.read_csv(path_b,sep=r"\s*",names=['item'],dtype=str)
dict_elts_total = {}
dict_elts_sizemin = {}
for c in classes:
pd_b[c] = 0
dict_elts_total[c] = 0
dict_elts_sizemin[c] = 0
for index, row in pd_b.iterrows():
i = row['item']
path_i = '/media/gonthier/HDD/data/Wikidata_Paintings/WikiTenLabels/Annotations/%s.xml'%(i)
read_file = voc_eval.parse_rec(path_i)
for element in read_file:
classe_elt = element['name']
bbox = element['bbox']
xmin = bbox[0]
ymin = bbox[1]
xmax = bbox[2]
ymax = bbox[3]
area = (xmax -xmin)*(ymax-ymin)
# print(area)
for c in classes:
if classe_elt==c: # We got an element from
pd_b.loc[pd_b['item']==row['item'],c] = 1
dict_elts_total[c] += 1
if area > size_min:
dict_elts_sizemin[c] += 1
print('Statistiques au niveaux du nombre de classes avec les labels dans la partie annotee en detection')
print(pd_b.sum())
print('Nombre d instances des differentes classes')
num_obj = 0
for c in classes:
print(c,' : ',dict_elts_total[c])
num_obj+=dict_elts_total[c]
print('Nombre d instances totales',num_obj)
print('Nombre d instances des differentes classes avec une taille superieur a :',size_min,'pixels',num_obj)
num_obj = 0
for c in classes:
print(c,' : ',dict_elts_sizemin[c])
num_obj+=dict_elts_sizemin[c]
print('Nombre d instances de taille superieur a ',size_min,'pixels',num_obj)
def CASPApaintings(copyFile=False,convertXML=False,copyIm=False):
"""
Creation of the CASPA paintings dataset for WSOD training and evaluation
"""
df = None
classes = ["bear", "bird", "cat", "cow", "dog", "elephant", "horse", "sheep"]
old_names = ["Bear", "bird", "cat", "Cattle Cow Bull", "dog", "elephant", "horse", "sheep shepherd"]
old_img_folder = 'Images_Per_Class'
default_path_imdb = '/media/gonthier/HDD2/data/'
if convertXML:
for old_name in old_names :
ff = old_name.lower()
ff = ff.replace(' ','_')
for folder in ['non','realistic']:
f = ff +'_'+folder
old_path= default_path_imdb + 'CASPApaintings/Annotations/'+f
list_xml = glob.glob(old_path+'/*.xml')
for elt in list_xml:
dst = default_path_imdb + 'CASPApaintings/Annotations/' + elt.split('/')[-1]
path_img = default_path_imdb + 'CASPApaintings/JPEGImages/' + elt.split('/')[-1].replace('xml','jpg')
# read the old xml file
old_tree=ET.parse(elt)
root = old_tree.getroot()
size = root.find('imagesize')
width = int(size.find('ncols').text)
height = int(size.find('nrows').text)
list_bd = []
for obj in root.iter('object'):
cls = obj.find('name').text
if cls not in classes:
continue
#cls_id = classes.index(cls)
polygon = obj.find('polygon')
points = polygon.iter('pt')
xmin = None
ymin = None
xmax = None
ymax = None
for i,pt in enumerate(points):
x = int(pt.find('x').text)
y = int(pt.find('y').text)
if i == 0:
xmin = x
ymin = y
elif i ==1:
xmax = x
elif i==2:
ymax = y
list_bd += [[cls,xmin,ymin,xmax,ymax]]
WriteNewXMLfile(dst,path_img,width,height,list_bd)
if copyIm:
# Copie of images in the new folders
for f_name in old_names:
for folder in ['non','realistic']:
old_path= default_path_imdb + 'CASPApaintings/' + old_img_folder +'/'+f_name+'/'+folder
list_im = glob.glob(old_path+'/*.jpg')
for elt in list_im:
lower_name_img = elt.split('/')[-1]
lower_name_img = lower_name_img.lower()
print(lower_name_img)
dst = default_path_imdb + 'CASPApaintings/JPEGImages/' + lower_name_img
copyfile(elt,dst)
if copyFile:
# Creation of the train and set file per class
df_test= None
df_train = None
for new_name,old_name in zip(classes,old_names):
#print(new_name,old_name)
path_c = default_path_imdb + 'CASPApaintings/' + old_img_folder+'/' + old_name + '.txt'
pd_c = pd.read_csv(path_c,names=['name_img'],dtype=str, encoding='utf-8')
pd_c_train, pd_c_test = train_test_split(pd_c,test_size=0.5, random_state=0)
pd_c_train['name_img'] = pd_c_train['name_img'].str.replace('.jpg','')
pd_c_train['name_img'] = pd_c_train['name_img'].str.lower()
# pd_c_train = pd_c_train.apply((lambda x : x.replace('.jpg','')),axis=0)
pd_c_test['name_img'] = pd_c_test['name_img'].str.replace('.jpg','')
pd_c_test['name_img'] = pd_c_test['name_img'].str.lower()
# pd_c_test =pd_c_test.apply((lambda x : x.replace('.jpg','')),axis=0)
path_c_train = default_path_imdb + 'CASPApaintings/ImageSets/Main/' + new_name + '_train.txt'
path_c_test = default_path_imdb + 'CASPApaintings/ImageSets/Main/' + new_name + '_test.txt'
pd_c_train.to_csv(path_c_train,header=False,index=False)
pd_c_test.to_csv(path_c_test,header=False,index=False)
#print(pd_c_test.head(5))
if df_train is None:
df_train = pd_c_train
else:
df_train = df_train.append(pd_c_train)
if df_test is None:
df_test = pd_c_test
else:
df_test = df_test.append(pd_c_test)
df_test = df_test.drop_duplicates(subset ='name_img')
df_train = df_train.drop_duplicates(subset ='name_img')
df_train = df_train['name_img']
df_test = df_test['name_img']
#print(df_test.head(5))
num_test_drop = 0
for row in df_test.values:
name_im = default_path_imdb + 'CASPApaintings/JPEGImages/' + row +'.jpg'
if not os.path.isfile(name_im):
#print(row,name_im)
df_test = df_test[df_test!=row]
num_test_drop += 1
num_train_drop = 0
for row in df_train.values:
name_im = default_path_imdb + 'CASPApaintings/JPEGImages/' + row +'.jpg'
if not os.path.isfile(name_im):
print(row,name_im)
df_train = df_train[df_train!=row]
num_train_drop += 1
print('number of drop images for train :',num_train_drop,'for test',num_test_drop)
path_train =default_path_imdb + 'CASPApaintings/ImageSets/Main/train.txt'
path_test =default_path_imdb + 'CASPApaintings/ImageSets/Main/test.txt'
df_train.to_csv(path_train,header=False,index=False)
df_test.to_csv(path_test,header=False,index=False)
# Copie of the xml files
# Besoin de convertir en Pascal voc format
oneNotFound = False
list_mising_xml = []
sets = [('CASPApaintings','train'),('CASPApaintings','test')]
for base,image_set in sets:
path_b = '/media/gonthier/HDD2/data/CASPApaintings/ImageSets/Main/%s.txt'%(image_set)
pd_b = pd.read_csv(path_b,names=['name_img'],dtype=str, encoding='utf-8')
for c in classes:
pd_b[c] = -1
for index, row in pd_b.iterrows():
i = row['name_img']
path_i = '/media/gonthier/HDD2/data/CASPApaintings/Annotations/%s.xml'%(i)
try:
read_file = voc_eval.parse_rec(path_i)
for element in read_file:
classe_elt = element['name']
for c in classes:
if classe_elt==c:
pd_b.loc[pd_b['name_img']==row['name_img'],c] = 1
except FileNotFoundError:
print(path_i,'not found')
oneNotFound = True
list_mising_xml += [i]
pd_b['set'] = image_set
if df is None:
df = pd_b
else:
df = df.append(pd_b)
df = df.drop_duplicates(subset ='name_img')
print('Size dataset :',len(df))
print(list_mising_xml)
# for i in list_mising_xml:
# name_img = default_path_imdb + 'CASPApaintings/JPEGImages/' + i +'.jpg'
# plt.figure()
# imreadoutput = plt.imread(name_img)
# plt.imshow(imreadoutput)
# input('wait')
c_missing = [['bear','horse'],['bear'],['bear'],['elephant'],['horse'],['horse'],['horse'],['horse'],['bear'],['bear','sheep'],['bear','dog'],['bear'],['bear'],['bear'],['horse']]
print('Modification of missing elements')
for classes,name in zip(c_missing,list_mising_xml):
print(name,df[df['name_img']==name]['set'])
for c in classes:
df.loc[df['name_img']==name,c] = 1
df.loc[df['name_img']==name,'set'] = 'train'
print('Size of datasets : ',len(df),'size train set :',len(df[df['set']=='train']),'size test set :',len(df[df['set']=='test']))
oneNotFound = False
df.loc[df['name_img']=='4-cat-and-cattle','cat'] = 1
if not(oneNotFound):
output_name = path_output + 'CASPApaintings_all' + '.csv'
df.to_csv(output_name,index=False)
output_name = path_output + 'CASPApaintings' + '.csv'
df.to_csv(output_name,index=False)
output_name = default_path_imdb + 'CASPApaintings/ImageSets/Main/CASPApaintings.csv'
df.to_csv(output_name,index=False)
else:
output_name = path_output + 'CASPApaintings_tmp' + '.csv'
df.to_csv(output_name,index=False)
print('XML not found, not saved')
# Il faut extraire les fichiers test et etc
output_name_csv = path_output + 'CASPApaintings' + '.csv'
print(output_name_csv)
df_new = pd.read_csv(output_name_csv,dtype=str)
for set in ['train','test']:
df_set = df_new[df_new['set']==set]
print(set,len(df_set))
all_names = df_set['name_img'].values
output_name = default_path_imdb + 'CASPApaintings/ImageSets/Main/'+set+'.txt'
np.savetxt(output_name, all_names, delimiter='\n',fmt='%s')
for c in classes:
df_c = df_set[df_set[c]==str(1)]
all_names_c = df_c['name_img'].values
output_name = default_path_imdb + 'CASPApaintings/ImageSets/Main/'+c+'_'+set+'.txt'
np.savetxt(output_name, all_names_c, delimiter='\n',fmt='%s')