def __init__(self,
select="all",
cl="face_train.txt",
basepath="media/DADES-2/",
trainfile="",
imagepath="afw/testimages/",
annpath="afw/testimages/",
local="afw/",
usetr=False,
usedf=False,
minh=0,
minw=0,
useOldAnn=False):
self.usetr = usetr
self.usedf = usedf
self.local = basepath + local
self.imagepath = basepath + imagepath
self.annpath = basepath + annpath
self.useOldAnn = useOldAnn
if useOldAnn:
self.trainfile = "annotations/anno2.mat"
self.ann = loadmat(self.trainfile)["anno"]
else:
self.trainfile = "annotations/new_annotations_AFW.mat"
self.ann = loadmat(self.trainfile)["Annotations"]
self.total = len(self.ann)
self.minh = minh
self.minw = minw
def __init__(
self,
select="all",
cl="face_train.txt",
basepath="media/DADES-2/",
trainfile="annotations/Annotations_Face_PASCALLayout_large_fixed.mat",
imagepath="/xxx/",
annpath="afw/testimages/",
local="afw/",
usetr=False,
usedf=False,
minh=0,
minw=0,
useOldAnn=False):
self.usetr = usetr
self.usedf = usedf
self.local = basepath + local
if useOldAnn:
pp = trainfile.find("_fixed")
self.trainfile = trainfile[:pp] + ".mat"
else:
self.trainfile = trainfile
self.imagepath = basepath + imagepath
self.annpath = basepath + annpath
self.minh = minh
self.minw = minw
self.ann = loadmat(self.trainfile)["Annotations"]
self.total = len(self.ann)
def similarity_matrix_from_file(embedding_file, vocabulary_file):
#df = pd.read_csv(embedding_file, index_col=0)
df = loadmat(embedding_file).toarray()
word_vectors = np.matrix(df)
df_vocab = np.ravel(np.matrix(pd.read_csv(vocabulary_file, index_col = 0)))
return similarity_matrix(word_vectors, df_vocab)
def loadDetectionsRamanan(fn):
f = util.loadmat(fn)
ids = f['ids']
scores = []
if f.has_key('sc'):
scores = f['sc']
boxes = f['BB']
n = len(ids)
det = []
for i in range(n):
this_id = ids[i][0][0].split(".")[0]
if not scores == []:
this_score = scores[i][0]
else:
this_score = 1.0
box = boxes[:, i]
x1 = float(box[0])
y1 = float(box[1])
x2 = float(box[2])
y2 = float(box[3])
det.append([this_id, this_score, x1, y1, x2, y2])
if 0:
im = util.myimread(
"/users/visics/mpederso/databases/afw/testimages/" + this_id + ".jpg")
pylab.clf()
pylab.imshow(im)
util.box([y1, x1, y2, x2])
pylab.draw()
pylab.show()
raw_input()
dets = sorted(det, key=itemgetter(1), reverse=True)
return dets
def similarity_matrix_coclustering_from_file(embedding_file, word_label_file, vocabulary_file):
#word_vectors = np.matrix(pd.read_csv(embedding_file, index_col=0))
word_vectors = np.matrix(loadmat(embedding_file).toarray())
label = pd.read_csv(word_label_file, index_col=0)
df_vocab = np.ravel(np.matrix(pd.read_csv(vocabulary_file, index_col = 0)))
return similarity_matrix_coclustering(word_vectors, label, df_vocab)
def loadDetectionsRamanan(fn):
f = util.loadmat(fn)
ids = f['ids']
scores = []
if f.has_key('sc'):
scores = f['sc']
boxes = f['BB']
n = len(ids)
det = []
for i in range(n):
this_id = ids[i][0][0].split(".")[0]
if not scores == []:
this_score = scores[i][0]
else:
this_score = 1.0
box = boxes[:, i]
x1 = float(box[0])
y1 = float(box[1])
x2 = float(box[2])
y2 = float(box[3])
det.append([this_id, this_score, x1, y1, x2, y2])
if 0:
im = util.myimread(
"/users/visics/mpederso/databases/afw/testimages/" + this_id +
".jpg")
pylab.clf()
pylab.imshow(im)
util.box([y1, x1, y2, x2])
pylab.draw()
pylab.show()
raw_input()
dets = sorted(det, key=itemgetter(1), reverse=True)
return dets
def loadDetectionsShen(fn):
f = util.loadmat(fn)
det = []
for idl, dd in enumerate(f["DetectionResults"]):
for ff in dd[0][0]["faces"][0]:
det.append([dd[0][0]["filename"][0][0].split(
"\\")[-1].split(".")[0], ff[4], ff[0], ff[1], ff[0] + ff[2], ff[1] + ff[3]])
dets = sorted(det, key=itemgetter(1), reverse=True)
return det
def loadDetectionsShen(fn):
f = util.loadmat(fn)
det = []
for idl, dd in enumerate(f["DetectionResults"]):
for ff in dd[0][0]["faces"][0]:
det.append([
dd[0][0]["filename"][0][0].split("\\")[-1].split(".")[0],
ff[4], ff[0], ff[1], ff[0] + ff[2], ff[1] + ff[3]
])
dets = sorted(det, key=itemgetter(1), reverse=True)
return det
def all_labels(labelFolder):
allFiles = glob(labelFolder + "/labels_*.mat")
filteredFiles = [x for x in allFiles if not "_esat" in x]
allLabels = []
for f in filteredFiles:
mat = util.loadmat(f)
for label in mat["labels_name"]:
allLabels.append(label + ",%s" % f)
return allLabels
def loadDetectionsYann(fn):
f = util.loadmat(fn)
det = []
widths = []
heights = []
size = f['ids'].shape[0]
bb = f['BB']
for i in range(size):
key = f['ids'][i][0][0].split('.')[0]
conf = float(f['confidence'][i][0])
if f.has_key("del"):
if f["del"][i] == 1:
continue
x1 = float(bb[0][i])
y1 = float(bb[1][i])
x2 = float(bb[2][i])
y2 = float(bb[3][i])
det.append([key, conf, x1, y1, x2, y2])
dets = sorted(det, key=itemgetter(1), reverse=True)
return dets
def loadDetectionsYann(fn):
f = util.loadmat(fn)
det = []
widths = []
heights = []
size = f['ids'].shape[0]
bb = f['BB']
for i in range(size):
key = f['ids'][i][0][0].split('.')[0]
conf = float(f['confidence'][i][0])
if f.has_key("del"):
if f["del"][i] == 1:
continue
x1 = float(bb[0][i])
y1 = float(bb[1][i])
x2 = float(bb[2][i])
y2 = float(bb[3][i])
det.append([key, conf, x1, y1, x2, y2])
dets = sorted(det, key=itemgetter(1), reverse=True)
return dets
def main():
print("Fetching labels")
labels = all_labels(labelFolder)
f = h5py.File(outPath, 'w')
depth = f.create_group("depth").create_dataset("depth_data",
(55, 74, len(labels)),
compression="lzf",
dtype=np.float64)
dt = h5py.special_dtype(vlen=bytes)
depth_label = f["depth"].create_dataset("depth_labels", [len(labels)],
dtype=dt)
depth_mat_source = f["depth"].create_dataset("depth_folder_id",
[len(labels)],
dtype=np.uint16)
allFiles = glob(labelFolder + "/labels_*.mat")
filteredFiles = [x for x in allFiles if not "_esat" in x]
startI = 0
fileNo = 1
# Loop over all data
for x in filteredFiles:
print("Processing file %d/%d" % (fileNo, len(filteredFiles)))
matFileNo = int(os.path.split(x)[-1].split("_")[1])
fdata = util.loadmat(x)
endI = startI + fdata["labels_name"].shape[0]
depth[:, :, startI:endI] = fdata["labels_processed"]
depth_label[startI:endI] = fdata["labels_name"]
depth_mat_source[startI:endI] = matFileNo
# Update startI
startI = endI
fileNo += 1
base_file = dataset_path+"/"+data_version+"/"+dataset
label_file = base_file+"_preprocessed.csv"
mat_file = base_file+"_preprocessed.mat"
embedding_file = base_file+"_preprocessed_embedding.mat"
vocab_file = base_file+"_preprocessed_vocabulary.csv"
df = pd.read_csv(label_file)
y = np.unique(df['Label'], return_inverse=True)[1] # as factor
mat = io.loadmat(mat_file)['X']
print(mat.shape)
no_cluster = len(np.unique(y))
print(no_cluster)
word_vectors = np.matrix(loadmat(embedding_file).toarray())
print(word_vectors.shape)
df_vocab = np.ravel(np.matrix(pd.read_csv(vocab_file, index_col = 0)))
algo_pipeline = []
algo_pipeline.append((CoclustInfo(n_row_clusters=no_cluster, n_col_clusters=no_cluster, n_init=10, max_iter=200), "CoclustInfo"))
algo_pipeline.append((CoclustMod(n_clusters=no_cluster, n_init=10, max_iter=200), "CoclustMod"))
algo_pipeline.append((CoclustSpecMod(n_clusters=no_cluster, n_init=10, max_iter=200), "CoclustSpecMod"))
for model, model_name in algo_pipeline:
res_nmi, res_ari, res_acc = execute_algo(model, model_name, mat, y)
res_accs, all_fp_words, all_fn_words = compute_column_metrics(model.column_labels_, word_vectors, df_vocab)
