def readLetters(self):
"""
Character recognition
"""
self.letters = []
i = 0
letters = []
for begin, end in self.segments:
lett = self.img[1 : self.line_height - 2, begin - 1 : end + 2] # add a bit of padding to help with
# recognition
if lett.shape[1] == 0:
continue
# self.img.drawRectangle(begin-1, 1, end - begin+2, 28)
if lett.shape[0] > lett.shape[1]:
lett = resize(lett, height=30)
else:
lett = resize(lett, width=30)
if lett.shape[0] > 0 and lett.shape[1] > 0:
lett = embiggen(lett, (30, 30))
# cv2.imwrite("tmp\\char %d %s.jpg" % (i, "r"), lett)
letters.append(lett.reshape(-1))
i += 1
try:
characters = labels.inverse_transform(logistic_model.predict(letters))
print characters
for charac, segment in zip(characters, self.segments):
self.letters.append((charac, segment[0], segment[1]))
except ValueError:
print (map(lambda x: x.shape, letters))
def get(self):
self.logger.info("Request to resize IMG from url")
name = "/tmp/rs-%s.png" % str(uuid.uuid4())
args = self._get_args()
resize(open_remote_image(self.get_query_argument("url")), name, args)
self.response_file(name)
def getComicMetadata(self, path):
ca = ComicArchive(path, default_image_path=AppFolders.missingPath("page.png"))
if ca.seemsToBeAComicArchive():
if ca.hasMetadata( MetaDataStyle.CIX ):
style = MetaDataStyle.CIX
elif ca.hasMetadata( MetaDataStyle.CBI ):
style = MetaDataStyle.CBI
elif ca.hasMetadata( MetaDataStyle.COMET ):
style = MetaDataStyle.COMET
elif ca.hasMetadata( MetaDataStyle.CBW ):
style = MetaDataStyle.CBW
else:
logging.warning(u"Library: File Has No ComicMeta Data")
if ca.hasMetadata( MetaDataStyle.CALIBRE ):
style = MetaDataStyle.CALIBRE
elif ca.hasMetadata( MetaDataStyle.EPUB ):
style = MetaDataStyle.EPUB
else:
style = None
if style is not None:
md = ca.readMetadata(style)
if md.isEmpty:
md = ca.metadataFromFilename()
else:
# No metadata in comic. make some guesses from the filename
md = ca.metadataFromFilename()
# patch version 2
if (md.title is None or md.title == "") and md.issue is None and not md.series is None:
md.title = md.series
md.series = None
md.fingerprint = ca.fingerprint()
md.path = ca.path
md.page_count = ca.page_count
md.mod_ts = datetime.utcfromtimestamp(getmtime(ca.path))
md.filesize = os.path.getsize(md.path)
md.hash = ""
#thumbnail generation
image_data = ca.getPage(0, AppFolders.missingPath("cover.png"))
#now resize it
thumb = StringIO.StringIO()
try:
utils.resize(image_data, (400, 400), thumb)
md.thumbnail = thumb.getvalue()
except:
md.thumbnail = None
return md
return None
def post(self):
self.logger.info("Request to resize IMG for request file")
args = self._get_args()
for item in self.request.files.values():
for file_info in item:
name = "/tmp/rs-%s-%s.png" % (time.time(), file_info["filename"])
resize(open_image(file_info["body"]), name, args)
self.response_file(name)
return
def save_pos_samples(root_folder, folder_to, train=True, subset_size=None):
"""Se encarga de cargar todos los samples positivos"""
lst_file_folder = 'Train' if train else 'Test'
lst_pos_file = os.path.join(lst_file_folder, 'pos.lst')
lst_annotations_file = os.path.join(lst_file_folder, 'annotations.lst')
content_pos = open(os.path.join(root_folder, lst_pos_file)) # Abro el listado de imagenes positivas
content_annotations = open(os.path.join(root_folder, lst_annotations_file)) # Abro el listado de imagenes positivas
content_pos_lines = content_pos.readlines()
content_annotations_lines = content_annotations.readlines()
# Si fue especificado un tamaño de subset recorto la lista de lineas
if subset_size:
combined = list(zip(content_pos_lines, content_annotations_lines))
# Los pongo en orden aleatorio cuando genero subset
random.shuffle(combined)
content_pos_lines, content_annotations_lines = zip(*combined)
# Recorto el numero de resultados
content_pos_lines = content_pos_lines[0:subset_size]
content_annotations_lines = content_annotations_lines[0:subset_size]
for img_path, bounding_boxes_path in zip(content_pos_lines, content_annotations_lines):
# Elimino el caracter de nueva linea
img_path = img_path.rstrip('\n')
bounding_boxes_path = bounding_boxes_path.rstrip('\n')
img_original = skimage.io.imread(os.path.join(root_folder, img_path)) # Cargo la imagen
# Obtengo los bounding boxes de personas a recortar
bounding_boxes_list = get_inria_bounding_boxes(root_folder, bounding_boxes_path)
for bounding_box in bounding_boxes_list:
persona = get_inria_bounding_box_cropped(img_original, bounding_box) # Recorto a la persona
persona = utils.resize(persona) # Re escalo la imagen
img_filename = utils.get_filename(img_path) # Genero el nombre que tendra la imagen guardada
utils.save_img(persona, folder_to, img_filename) # Guardo la imagen en la carpeta de positivos
def transform(img): # # [TITLE ONLY] # # standardize the size of imput image. extract ONLY the title of the poster. # h,w = img.shape[:2] # imgT = img[:h/5,:] # crop out the title # imgT = utils.resize(imgT, CFG.STD_SIZE) # h,w = imgT.shape[:2] # title = [(0,0), (w-1,0), (w-1,h-1), (0,h-1)] # coords of the title is that of the img itself # standardize the size of imput image imgT = utils.resize(img, CFG.STD_SIZE) imgT = pTr.addOcclusions(imgT) h,w = imgT.shape[:2] r = CFG.TITLE_RATIO title = [(0,0), (w-1,0), (w-1,int(h*r)), (0,int(h*r))] imgT, title = pTr.lightBlob(imgT, title) imgT, title = pTr.blur(imgT, title) imgT, title = pTr.scaleAndTranslate(imgT,title) imgT, title = pTr.perspective(imgT, title) imgT, title = pTr.rotate(imgT,title) titleArea = utils.boundingArea(title) tBox = utils.formatLabel(titleArea, imgT.shape[:2]) return (imgT, tBox)
def detect_edges(image): image = utils.resize(image, height = 500) gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) gray = cv2.GaussianBlur(gray, (5, 5), 0) edges = utils.auto_canny(gray) return edges
def upload_url(request):
albumform = PickAlbumForm(request.user, request.POST)
urlform = UploadFromURLForm(request.POST)
if albumform.is_valid() and urlform.is_valid():
url = urlform.cleaned_data['url']
album = albumform.cleaned_data['album']
name = urlparse(url).path.split('/')[-1]
tmp_img = NamedTemporaryFile(delete=True)
tmp_img.write(urllib2.urlopen(url).read())
tmp_img.flush()
max_order = Photo.objects.filter(album=album).aggregate(Max('order'))['order__max'] or 0
path = 'albums/%s/%s/' % (request.user.id, album.id)
photo = Photo(user=request.user, album=album)
photo.image.save(name, File(tmp_img))
photo.image.open()
# get the resizing dimensions from the preferences #TODO this might move to utils in the future
preferences = Preferences.get_or_create(request.user)
resize_dimensions = preferences.get_default_img_size()
img_data = resize(photo.image, upload_to=path, sizes_data=[resize_dimensions], overwrite=True)
for data in img_data:
photo.width=data[1]
photo.height=data[2]
photo.order = max_order + 1
photo.save()
p_id = photo.id
return HttpResponse(p_id, mimetype="text/plain")
return render(request, 'albums/uploadify_url.html', {'urlform': urlform})
def uploadify(request):
# Processing of each uploaded image
albumform = PickAlbumForm(request.user, request.POST)
# import pdb; pdb.set_trace()
if albumform.is_valid():
album = albumform.cleaned_data['album']
max_order = Photo.objects.filter(album=album).aggregate(Max('order'))['order__max'] or 0
img = request.FILES['Filedata']
path = 'albums/%s/%s/' % (request.user.id, album.id) #/media/albums/<userid>/<albumid>/<img>.jpg
# get the resizing dimensions from the preferences #TODO this might move to utils in the future
preferences = Preferences.get_or_create(request.user)
resize_dimensions = preferences.get_default_img_size()
img_data = resize(img, upload_to=path, sizes_data=[resize_dimensions])
# if img_data is not None:
for data in img_data:
photo = Photo(user=request.user, album=album, width=data[1], height=data[2])
photo.image = data[0]
photo.order = max_order + 1
photo.save()
p_id = photo.id
return HttpResponse('%s' % p_id, mimetype="text/plain") #return the photo id
# else:
# return HttpResponse(_('File extension not in valid extensions: "%(extensions)s".' % {
# 'extensions': ",".join(settings.VALID_IMG_EXTENSIONS)
# })
# )
else:
return HttpResponse()
def getValidation(self, params):
if self.validation_dict == None:
images = []
labels = []
# Read files
for sample_filepath in self.validation_list:
sample_fullpath = self.corpus_dirpath + '/' + sample_filepath + '/' + sample_filepath
# IMAGE
sample_img = cv2.imread(sample_fullpath + '.png',
False) # Grayscale is assumed!
height = params['img_height']
sample_img = utils.resize(sample_img, height)
images.append(utils.normalize(sample_img))
# GROUND TRUTH
if self.semantic:
sample_full_filepath = sample_fullpath + '.semantic'
else:
sample_full_filepath = sample_fullpath + '.agnostic'
sample_gt_file = open(sample_full_filepath, 'r')
sample_gt_plain = sample_gt_file.readline().rstrip().split(
utils.word_separator())
sample_gt_file.close()
labels.append([self.word2int[lab] for lab in sample_gt_plain])
# Transform to batch
image_widths = [img.shape[1] for img in images]
max_image_width = max(image_widths)
batch_images = np.ones(shape=[
len(self.validation_list), params['img_height'],
max_image_width, params['img_channels']
],
dtype=np.float32) * self.PAD_COLUMN
for i, img in enumerate(images):
batch_images[i, 0:img.shape[0], 0:img.shape[1], 0] = img
# LENGTH
width_reduction = 1
for i in range(params['conv_blocks']):
width_reduction = width_reduction * params[
'conv_pooling_size'][i][1]
lengths = [batch_images.shape[2] / width_reduction
] * batch_images.shape[0]
self.validation_dict = {
'inputs': batch_images,
'seq_lengths': np.asarray(lengths),
'targets': labels,
}
return self.validation_dict, len(self.validation_list)
def paste_pose(background):
# (75, 225) (243, 444)
img_path = random.choice(pose_list)
image = Image.open(img_path)
image = resize(image, 166, 219)
background.paste(image, (75-35, 225-38))
def images_align():
image = nib.load('20121114_075953s982300a000.nii.gz')
affine = image.affine
data = np.array(image.dataobj[:, :, :, 2])
image_nib = nib.Nifti1Image(data, affine=affine)
new_image_nib = resize(image_nib, (96, 96, 106), interpolation="nearest")
nib.save(new_image_nib, 'test.nii')
print(new_image_nib)
def pred_func(objects, last_output, gt_size):
obj_pred = last_output[0].clone()
for pred, o in zip(last_output[1:], objects):
obj_pred[:, o] = pred[:, o]
obj_pred = utils.resize(obj_pred, gt_size)
obj_pred = obj_pred.argmax(dim=1)
return obj_pred, None
def extract(path):
img = cv2.imread(path, cv2.IMREAD_COLOR)
img = resize(img, 300)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
faces = face_cascade.detectMultiScale(gray, 1.1, 3)
if len(faces) <= 0:
return None
bodies = body_cascade.detectMultiScale(gray, 1.1, 3)
if len(bodies) <= 0:
return None
face = Rectangle(faces[0][0], faces[0][1], faces[0][2], faces[0][3])
body = Rectangle(bodies[0][0], bodies[0][1], bodies[0][2], bodies[0][3])
# Body/Face Section
body_ratio = body.w / face.w
if body_ratio <= BODY_RATIO_THRESHOLD:
return None
# Skin Section
skinR = skinG = skinB = 0
for col in range(face.half_w):
for row in range(10):
pixel = img[face.y+face.half_h-row+5, face.x+face.quarter_w+col]
skinB += pixel[0]
skinG += pixel[1]
skinR += pixel[2]
averageDivisor = face.half_w * 10
skinR = int(skinR / averageDivisor)
skinG = int(skinG / averageDivisor)
skinB = int(skinB / averageDivisor)
# Hair Section
hairR = hairG = hairB = 0
top = face.y - 5
if top < 0:
top = 0
for col in range(face.half_w):
for row in range(10):
pixel = img[top+row, face.x+face.quarter_w+col]
hairB += pixel[0]
hairG += pixel[1]
hairR += pixel[2]
hairR = int(hairR / averageDivisor)
hairG = int(hairG / averageDivisor)
hairB = int(hairB / averageDivisor)
# Contour Section
contour = recognize_contour(img[face.y:face.y2, face.x:face.x2])
print("{}\t{}\t{}\t{}\t{}\t{}\t{}\t{}\t{}".format(os.path.basename(path), skinR, skinG, skinB, hairR, hairG, hairB, body_ratio, contour))
def get_next_batch(batch_size=128):
images = []
to_images = []
max_width_image = 0
font_min_length = random.randint(10, 20)
for i in range(batch_size):
font_name = random.choice(AllFontNames)
font_length = random.randint(font_min_length-5, font_min_length+5)
font_size = random.randint(9, 64)
font_mode = random.choice([0,1,2,4])
font_hint = random.choice([0,1,2,3,4,5])
text = utils_font.get_random_text(CHARS, eng_world_list, font_length)
image = utils_font.get_font_image_from_url(text, font_name ,font_size, fontmode = font_mode, fonthint = font_hint )
image = utils_font.add_noise(image)
image = utils_pil.convert_to_gray(image)
rate = random.randint(8, 17) / font_size
image = utils_pil.resize(image, rate)
image = np.asarray(image)
image = utils.resize(image, height=image_height)
image = 255. - image
images.append(image)
to_image = utils_font.get_font_image_from_url(text, font_name ,image_height, fontmode = font_mode, fonthint = font_hint)
to_image = utils_pil.convert_to_gray(to_image)
to_image = np.asarray(to_image)
to_image = utils.resize(to_image, height=image_height)
to_image = utils.img2bwinv(to_image)
to_images.append(to_image)
if image.shape[1] > max_width_image:
max_width_image = image.shape[1]
if to_image.shape[1] > max_width_image:
max_width_image = to_image.shape[1]
max_width_image = max_width_image + (POOL_SIZE - max_width_image % POOL_SIZE)
inputs = np.zeros([batch_size, max_width_image, image_height])
for i in range(len(images)):
image_vec = utils.img2vec(images[i], height=image_height, width=max_width_image, flatten=False)
inputs[i,:] = np.transpose(image_vec)
labels = np.zeros([batch_size, max_width_image, image_height])
for i in range(len(to_images)):
image_vec = utils.img2vec(to_images[i], height=image_height, width=max_width_image, flatten=False)
labels[i,:] = np.transpose(image_vec)
return inputs, labels
def resize_and_crop(phase, src, scale_size, crop_size):
# resize the images
src = resize(src, scale_size)
# crop the images
crop_params = get_crop_params(phase, src, crop_size)
src = crop(src, crop_params[0], crop_params[1], crop_params[2], crop_params[3])
return src
def face_verify():
start = time.time()
ensure_folder('static')
file1 = request.files['file1']
fn_1 = secure_filename(file1.filename)
full_path_1 = os.path.join('static', fn_1)
file1.save(full_path_1)
resize(full_path_1)
file2 = request.files['file2']
fn_2 = secure_filename(file2.filename)
full_path_2 = os.path.join('static', fn_2)
file2.save(full_path_2)
resize(full_path_2)
prob, is_same = compare(full_path_1, full_path_2)
elapsed = time.time() - start
return is_same, prob, elapsed, fn_1, fn_2
def process_scans(self, image):
image = segmentation_algo.get_slices_with_nodules(image)
image = utils.resize(image)
if self._augment:
angle = randrange(-15, 15)
image = utils.rotate_scans(image, angle)
return utils.trim_pad_slices(image, pad_with_existing=True)
def after_resize(bg_size, card, coordinates):
min_size=100
left, top, right, bottom = card.getbbox()
bw, bh = bg_size
if bw < right*bh/bottom:
new_size = random.randint(min_size, bw)
card = resize(card, width=new_size)
else:
new_size = random.randint(min_size, bh)
card = resize(card, height=new_size)
nw, nh = card.size
coordinates = coordinates.astype('float')
coordinates[:, 0] *= nw/right
coordinates[:, 1] *= nh/bottom
return card, coordinates
def copy_image(download_dir, overall_index, output_dir, args, i):
img_path = get_img_path(i, download_dir)
output_path = get_img_path(overall_index + i, output_dir)
if args.resize_to is None:
shutil.move(img_path, output_path)
else:
img = load_exr(img_path)
resized = resize(img, args.resize_to)
write_exr(output_path, resized)
def build_encoders(self, image):
ycbcr = utils.RGB2YCbCr(image)
encoders = []
M, N = image.shape[:2]
for i in range(3):
channel = ycbcr[:, :, i] if i == 0 else utils.resize(
ycbcr[:, :, i], M // 2, N // 2)
encoders.append(ZeroTreeEncoder(channel, WAVELET))
return encoders
def readImage():
img = cv2.imread('frame/ff206.png')
img = resize(img,1500)
im_blank = np.zeros_like(img)
preProcess(img)
imContours = img.copy()
getContours(imThres)
stack = imContours
cv2.imshow("stack", stack)
cv2.waitKey(20*1000)
def match_image():
start = time.time()
ensure_folder(STATIC_DIR)
ensure_folder(UPLOAD_DIR)
file1 = request.files['file1']
fn_1 = secure_filename(file1.filename)
full_path_1 = os.path.join(UPLOAD_DIR, fn_1)
file1.save(full_path_1)
resize(full_path_1)
file2 = request.files['file2']
fn_2 = secure_filename(file2.filename)
full_path_2 = os.path.join(UPLOAD_DIR, fn_2)
file2.save(full_path_2)
resize(full_path_2)
is_match = compare(full_path_1, full_path_2)
elapsed = time.time() - start
return is_match, elapsed, fn_1, fn_2
def FEtimesampling(self, wf):
#first time sampling:
step = float(1./fesampling)
tracelength = wf.length
nrofpoints = int(tracelength/step)
newtime = np.linspace(wf.tstart,wf.tend,nrofpoints)
[a,b] = utils.resize(wf.time,wf.amp)
newy = np.interp(newtime,a,b)
newwf = waveform.Waveform(newtime,newy,'timesampled')
return newwf
def FEtimesampling(self, wf):
#first time sampling:
step = float(1./fesampling)
tracelength = wf.length
nrofpoints = int(tracelength/step)
newtime = np.linspace(wf.tstart,wf.tend,nrofpoints)
[a,b] = utils.resize(wf.time,wf.amp)
newy = np.interp(newtime,a,b)
newwf = waveform.Waveform(newtime,newy,'timesampled')
return newwf
def scan(file):
image = Image.open(file.stream)
image = utils_pil.convert_to_gray(image)
image = np.asarray(image)
utils.save(image, os.path.join(curr_dir, "test", "p0.png"))
# image = utils.clearImgGray(image)
# utils.save(image * 255, os.path.join(curr_dir,"test","p1.png"))
split_images = utils.splitImg(image)
ocr_texts = []
for i, split_image in enumerate(split_images):
inv_image = utils.img2bwinv(split_image)
inv_image = utils.clearImg(inv_image)
image = 255. - split_image
image = utils.dropZeroEdges(inv_image, image)
image = utils.resize(image, ocr.image_height)
image = image / 255.
ocr_inputs = np.zeros([1, ocr.image_size, ocr.image_size])
ocr_inputs[0, :] = utils.square_img(
image, np.zeros([ocr.image_size, ocr.image_size]))
utils.save(ocr_inputs[0] * 255,
os.path.join(curr_dir, "test", "ocr_%s.png" % i))
ocr_seq_len = np.ones(1) * ocr.SEQ_LENGHT
start = time.time()
# p_net_g = session.run(net_g, {inputs: ocr_inputs})
# p_net_g = np.squeeze(p_net_g, axis=3)
# debug_net_g = np.copy(p_net_g)
# for j in range(1):
# _t_img = utils.unsquare_img(p_net_g[j], ocr.image_height)
# _t_img[_t_img<0] = 0
# _t_img = utils.cvTrimImage(_t_img)
# _t_img = utils.resize(_t_img, ocr.image_height)
# if _t_img.shape[0] * _t_img.shape[1] <= ocr.image_size * ocr.image_size:
# p_net_g[j] = utils.square_img(_t_img, np.zeros([ocr.image_size, ocr.image_size]), ocr.image_height)
# _img = np.vstack((ocr_inputs[0], debug_net_g[0], p_net_g[0]))
# utils.save(_img * 255, os.path.join(curr_dir,"test","%s.png"%i))
decoded_list = session.run(res_decoded[0], {
inputs: ocr_inputs,
seq_len: ocr_seq_len
})
seconds = round(time.time() - start, 2)
print("filished ocr %s , paid %s seconds" % (i, seconds))
detected_list = utils.decode_sparse_tensor(decoded_list)
for detect_number in detected_list:
ocr_texts.append(ocr.list_to_chars(detect_number))
return ocr_texts
def load_stylegan_avatar():
url = "https://thispersondoesnotexist.com/image"
r = requests.get(url, headers={'User-Agent': "Impersonator 0.1.0"}).content
image = np.frombuffer(r, np.uint8)
image = cv2.imdecode(image, cv2.IMREAD_COLOR)
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
image = resize(image, (IMG_SIZE, IMG_SIZE))
return image
def evaluate(track):
mix_audio, orig_sr, mix_channels = track.audio, track.rate, track.audio.shape[
1]
if mix_channels > 1:
mono_audio = librosa.to_mono(mix_audio.T)
else:
mono_audio = mix_audio
mono_audio = pad_audio(mono_audio, orig_sr)
if orig_sr != hp.sample_rate:
mono_audio = librosa.resample(mono_audio, orig_sr, hp.sample_rate)
estimates = model.generate_wav(device, mono_audio)
if hp.sample_rate != orig_sr:
resample(estimates, orig_sr)
resize(estimates, mix_audio)
if mix_channels > 1:
replicate_channels(estimates, mix_channels)
#scores = museval.eval_mus_track(
# track, estimates, output_dir='bss_evals')
#print(scores)
return estimates
def __call__(self, x_batch, y_batch):
images, landmarks = [], []
for img_file, pts_file in zip(x_batch, y_batch):
img = imread(img_file)
xs, ys = read_landmarks(pts_file)
pts = np.r_[xs, ys]
cropped = crop(img, pts, padding=self.padding)
new_img, new_pts = resize(*cropped, target_size=self.target_size)
images.append(new_img)
landmarks.append(new_pts)
return np.array(images), np.array(landmarks)
def paste_image(background_image, text, kind, space_width=3, height=25):
image, mask = _generate_horizontal_text(text, font_size=32, space_width=space_width)
image = resize(image, height=height)
idw = image.size[0]
x = (positions[kind][1][0] + positions[kind][0][0] - idw) // 2 + random.randint(-10, 10)
y = positions[kind][0][1]
background_image.paste(image, (x, y), image)
return image
def match_video():
start = time.time()
ensure_folder(STATIC_DIR)
ensure_folder(UPLOAD_DIR)
file = request.files['file']
upload_file = secure_filename(file.filename)
full_path = os.path.join(UPLOAD_DIR, upload_file)
file.save(full_path)
resize(full_path)
print('full_path: ' + full_path)
with torch.no_grad():
x = gen_feature(full_path)
cosine = np.dot(features, x)
cosine = np.clip(cosine, -1, 1)
print('cosine.shape: ' + str(cosine.shape))
max_index = int(np.argmax(cosine))
max_value = cosine[max_index]
name = name_list[max_index]
fps = fps_list[max_index]
idx = idx_list[max_index]
image_fn = image_fn_list[max_index]
print('max_index: ' + str(max_index))
print('max_value: ' + str(max_value))
print('name: ' + name)
print('fps: ' + str(fps))
print('idx: ' + str(idx))
theta = math.acos(max_value)
theta = theta * 180 / math.pi
print('theta: ' + str(theta))
prob = get_prob(theta)
print('prob: ' + str(prob))
time_in_video = idx / fps
print('time_in_video: ' + str(time_in_video))
prob = get_prob(theta)
elapsed = time.time() - start
return name, prob, idx, float(time_in_video), float(elapsed), str(
upload_file), image_fn
def __call__(self, pred, obj_gt, part_gt):
obj_pred, part_pred = pred
obj_gt = obj_gt.squeeze(dim=1)
part_gt = part_gt.squeeze(dim=1)
pred_size = obj_pred.shape[-2:]
obj_gt = utils.resize(obj_gt, pred_size)
part_gt = utils.resize(part_gt, pred_size)
part_gt = self.tree.split_parts_gt(obj_gt, part_gt, mark_in_obj=False)
obj_loss = self.obj_ce(obj_pred, obj_gt)
part_loss = []
for o, (start, end) in self.tree.obj2part_idx.items():
obj_mask = obj_gt == o
o_part_gt = part_gt[self.tree.obj2idx[o]].clone()
o_part_gt[~obj_mask] = -1
o_part_pred = part_pred[:, start:end]
part_loss.append(self.part_ce(o_part_pred, o_part_gt))
loss = obj_loss + sum(part_loss)
return loss
def descriptors_from_class(dataset,
class_img_paths,
class_number,
option=constants.ORB_FEAT_OPTION):
"""
Gets all the local descriptors for a class. If an image has a side with more than 640 pixels it will be resized
leaving the biggest side at 640 pixels and conserving the aspect ratio for the other side.
Args:
dataset (Dataset object): An object that stores information about the dataset.
class_img_paths (array of strings): The paths for each image in certain class.
class_number (integer): The number of the class.
option (integer): If this is 49 (The key '1') uses ORB features, else use SIFT.
Returns:
numpy float matrix: Each row are the descriptors found in an image of the class
"""
des = None
#print class_img_paths
step = (constants.STEP_PERCENTAGE * len(class_img_paths)) / 100
for i in range(len(class_img_paths)):
img_path = class_img_paths[i]
img = cv2.imread(img_path)
resize_to = 640
h, w, channels = img.shape
if h > resize_to or w > resize_to:
img = utils.resize(img, resize_to, h, w)
if option == constants.ORB_FEAT_OPTION:
des_name = "ORB"
new_des = orb(img)
else:
des_name = "SIFT"
new_des = sift(img)
if new_des is not None:
if des is None:
des = nmpy.array(new_des, dtype=nmpy.float32)
#print "YES"
else:
des = nmpy.vstack((des, nmpy.array(new_des)))
#print "NO"
# Print a message to show the status of the function
if i % step == 0:
percentage = (100 * i) / len(class_img_paths)
message = "Calculated {0} descriptors for image {1} of {2}({3}%) of class number {4} ...".format(
des_name, i, len(class_img_paths), percentage, class_number)
print(message)
message = "* Finished getting the descriptors for the class number {0}*".format(
class_number)
print(message)
print("Number of descriptors in class: {0}".format(len(des)))
dataset.set_class_count(class_number, len(des))
return des
def load_mask(self, img, index):
imgh, imgw = img.shape[0:2]
mask_type = self.mask
# external + random block
if mask_type == 4:
mask_type = 1 if np.random.binomial(1, 0.5) == 1 else 3
# external + random block + half
elif mask_type == 5:
mask_type = np.random.randint(1, 4)
# random block
if mask_type == 1:
# return create_mask(imgw, imgh, imgw // 2, imgh // 2)
return mask_generator.generate_random_mask(
np.random.rand(np.random.randint(2, 5), 8), imgh, imgw)
# half
if mask_type == 2:
# randomly choose right or left
return create_mask(imgw, imgh, imgw // 2, imgh,
0 if random.random() < 0.5 else imgw // 2, 0)
# external
if mask_type == 3:
mask_index = random.randint(0, len(self.mask_data) - 1)
mask = imread(self.mask_data[mask_index])
mask = utils.resize(mask, imgh, imgw, self.is_center_crop)
mask = (mask > 0).astype(
np.uint8) * 255 # threshold due to interpolation
return mask
# test mode: load mask non random
if mask_type == 6:
mask = imread(self.mask_data[index % len(self.mask_data)])
mask = utils.resize(mask, imgh, imgw, centerCrop=False)
mask = rgb2gray(mask)
mask = (mask > 0).astype(np.uint8) * 255
return mask
def straighten(self, stepsize=3, low_angle=-5, high_angle=5):
# width is shape[1], height is shape[0]
#cv2.imwrite("tmp//before.jpg", self.img)
origWidth = self.img.shape[1]
if low_angle < -20 or high_angle > 20:
raise Exception("Photo is too skewed. Please straighten photo before trying to process it")
img = resize(self.img, width=600) # for some reason, straightening works better at this width :-??
# straighten out images
# using histograms: rotate +-5 in .3 steps, get max of each histogram
# and ideal rotation is argmax of those maxes
simg = img
hists = []
rng = list(range(low_angle*stepsize, high_angle*stepsize))
bincount = 600 if img.shape[0] > 600 else img.shape[0]
for ang in rng:
pimg = rotate(simg, ang/float(stepsize), fixed=True) # was true, but doesn't make sense and doens't work
# pimg = rotate(simg, ang/float(stepsize))
# cv2.imwrite("tmp//rotate %d.jpg" % ang, pimg)
hist, _ = np.histogram(pimg.sum(axis=1), bincount)
#plt.plot(hist)
#plt.savefig('tmp//hist %d.png' % ang, bbox_inches='tight')
#plt.close()
hists.append(max(hist))
rot = np.argmax(hists)
# if the best rotation angle is the one on the edge of our threshold, try to rotate again with an extended
# threshold in that direction
if rot == 0:
self.straighten(low_angle=low_angle-5, high_angle=high_angle-5)
elif rot == len(rng) - 1:
self.straighten(low_angle=low_angle+5, high_angle=high_angle+5)
img = rotate(self.img, rng[rot]/float(stepsize), fixed=False) # otsu's method removes
# background noise better
# self.img = img.resize(w=origWidth//2) # so that all letters are small enough
self.img = resize(img, width=600) # maybe I should look at average size of a blob ?
def left_blending(left_img, warped_img, disx, disy):
points_left = approx(left_img)
left_y, left_x = check_border(points_left)
points_warped = approx(warped_img)
warped_y, warped_x = check_border(points_warped)
min_y_a = min(left_y)
min_x_a = min(left_x)
max_x_a = max(left_x)
max_x = max(warped_x)
max_y = max(warped_y)
# 左图
A = left_img[0:max_x_a, min_y_a:disx, :].copy()
A0 = left_img[disy + 200:disy + 300, disx:disx + 150, :] # 光照补偿块
# 右图
B = np.zeros((A.shape[0], max_y, 3), np.uint8)
B[disy:disy + max_x, 0:max_y, :] = warped_img[0:max_x, 0:max_y, :]
# 右图
B0 = B[disy + 200:disy + 300, 0:150, :] # 光照补偿参照块
k = xi(A0, B0)
left_img = Recorver(left_img, k)
#import sys
#sys.exit()
A_shape = A.shape[:2]
B_shape = B.shape[:2]
padding = 8
if padding % 2 != 0:
padding = padding + 1
A1 = left_img[0:max_x_a, min_y_a:disx + 2 * padding, :].copy()
std_size = resize(A_shape, B_shape, padding * 2)
#print(std_size) # 标准大小
left_padding = img_padding(A1, std_size, padding, bi='left')
right_padding = img_padding(B, std_size, padding, bi='right')
#plt.imshow(left_padding)
#plt.show()
#plt.imshow(right_padding)
#plt.show()
result = multi_band_blending(left_padding, right_padding, padding * 2, 3,
False)
#print(left_padding.shape,right_padding.shape,result.shape)
#plt.imshow(result)
#plt.show()
re = left_padding.shape[1] - A.shape[1]
result1 = result[min_x_a:, re:, :].copy()
return result1
def prepare_x(self, input_directory, do_resize=True):
data_dir = pathlib.Path(input_directory)
input_images_url = list(data_dir.glob('*'))
if do_resize:
train_x = np.asarray(
[np.array(to_grayscale(resize(PIL.Image.open(str(image_url))))) for image_url in input_images_url])
print(train_x.shape)
train_x = np.expand_dims(train_x, -1)
else:
train_x = [np.array(to_grayscale(PIL.Image.open(str(image_url)))) for image_url in input_images_url]
print("#Images:", len(input_images_url))
print("train_x ready")
return train_x
def load_item(self, index):
size = self.input_size
# load image
ori_img = self.data[index]
img = np.array(ori_img)
# gray to rgb
if len(ori_img.shape) < 3:
ori_img = gray2rgb(ori_img)
# resize/crop if needed
if size != 0:
img = utils.resize(ori_img, size, size, self.is_center_crop)
# create grayscale image
img_gray = rgb2gray(img)
# load mask
mask = self.load_mask(img, index)
# resize/crop if needed
if size != 0:
img_shape = img.shape
mask = utils.resize(mask, img_shape[0], img_shape[1],
self.is_center_crop)
# load edge
edge = self.load_edge(img_gray, index, mask)
# augment data
if self.augment and np.random.binomial(1, 0.5) > 0:
img = img[:, ::-1, ...]
img_gray = img_gray[:, ::-1, ...]
edge = edge[:, ::-1, ...]
mask = mask[:, ::-1, ...]
return list(ori_img.shape), self.to_tensor(img), self.to_tensor(
img_gray), self.to_tensor(edge), self.to_tensor(mask)
def test(src_cam: cv.VideoCapture,
dst_cam: cv.VideoCapture,
affine_mat,
size=(640, 480)):
while True:
# read images from both cameras
is_ok, src_image = src_cam.read()
if not is_ok:
continue
is_ok, dst_image = dst_cam.read()
if not is_ok:
continue
# resize
src_resized = utils.resize(src_image, size)
dst_resized = utils.resize(dst_image, size)
# transform
src_resized = cv.warpAffine(src_resized,
affine_mat,
dst_resized.shape[::-1],
flags=cv.INTER_LINEAR)
# get BGR and Gray
src_bgr, src_gray = utils.to_bgr_gray(src_resized)
dst_bgr, dst_gray = utils.to_bgr_gray(dst_resized)
# combine
combine = (0.5 * src_bgr + 0.5 * dst_bgr).astype(np.uint8)
# draw
cv.imshow('combine', combine)
key = cv.waitKey(10)
if key == 27:
break
cv.destroyAllWindows()
def getComicMetadata(self, path):
ca = ComicArchive(
path, default_image_path=AppFolders.imagePath("default.jpg"))
if ca.seemsToBeAComicArchive():
logging.debug(u"Reading in {0} {1}\r".format(
self.read_count, path))
sys.stdout.flush()
self.read_count += 1
if ca.hasMetadata(MetaDataStyle.CIX):
style = MetaDataStyle.CIX
elif ca.hasMetadata(MetaDataStyle.CBI):
style = MetaDataStyle.CBI
else:
style = None
if style is not None:
md = ca.readMetadata(style)
else:
# No metadata in comic. make some guesses from the filename
md = ca.metadataFromFilename()
md.path = ca.path
md.page_count = ca.page_count
md.mod_ts = datetime.utcfromtimestamp(os.path.getmtime(ca.path))
md.filesize = os.path.getsize(md.path)
md.hash = ""
#thumbnail generation
image_data = ca.getPage(0)
#now resize it
thumb = StringIO.StringIO()
utils.resize(image_data, (200, 200), thumb)
md.thumbnail = thumb.getvalue()
return md
return None
def get_shrunk_channels(self, src):
shrink = self.options["shrink"]
n_orient = self.options["n_orient"]
grd_smooth_rad = self.options["grd_smooth_rad"]
grd_norm_rad = self.options["grd_norm_rad"]
luv = rgb2luv(src)
size = (luv.shape[0] / shrink, luv.shape[1] / shrink)
channels = [resize(luv, size)]
for scale in [1.0, 0.5]:
img = resize(luv, (luv.shape[0] * scale, luv.shape[1] * scale))
img = conv_tri(img, grd_smooth_rad)
magnitude, orientation = gradient(img, grd_norm_rad)
downscale = max(1, int(shrink * scale))
hist = histogram(magnitude, orientation, downscale, n_orient)
channels.append(resize(magnitude, size)[:, :, None])
channels.append(resize(hist, size))
channels = N.concatenate(channels, axis=2)
reg_smooth_rad = self.options["reg_smooth_rad"] / float(shrink)
ss_smooth_rad = self.options["ss_smooth_rad"] / float(shrink)
if reg_smooth_rad > 1.0:
reg_ch = conv_tri(channels, int(round(reg_smooth_rad)))
else:
reg_ch = conv_tri(channels, reg_smooth_rad)
if ss_smooth_rad > 1.0:
ss_ch = conv_tri(channels, int(round(ss_smooth_rad)))
else:
ss_ch = conv_tri(channels, ss_smooth_rad)
return reg_ch, ss_ch
def getComicMetadata(self, path):
ca = ComicArchive(path, default_image_path=AppFolders.imagePath("default.jpg"))
if ca.seemsToBeAComicArchive():
logging.debug(u"Reading in {0} {1}\r".format(self.read_count, path))
sys.stdout.flush()
self.read_count += 1
if ca.hasMetadata( MetaDataStyle.CIX ):
style = MetaDataStyle.CIX
elif ca.hasMetadata( MetaDataStyle.CBI ):
style = MetaDataStyle.CBI
else:
style = None
if style is not None:
md = ca.readMetadata(style)
else:
# No metadata in comic. make some guesses from the filename
md = ca.metadataFromFilename()
md.path = ca.path
md.page_count = ca.page_count
md.mod_ts = datetime.utcfromtimestamp(os.path.getmtime(ca.path))
md.filesize = os.path.getsize(md.path)
md.hash = ""
#thumbnail generation
image_data = ca.getPage(0)
#now resize it
thumb = StringIO.StringIO()
utils.resize(image_data, (200, 200), thumb)
md.thumbnail = thumb.getvalue()
return md
return None
def upload(request):
"""
Pure HTTP based upload, if you don't want to or can't use Flash and/or javascript.
"""
amountform = AmountForm(request.GET)
if amountform.is_valid():
amount = amountform.cleaned_data['amount'] or 20
else:
amount = 20
PhotoFormSet = modelformset_factory(Photo, fields=('image',), extra=amount)
if request.method == "POST":
albumform = PickAlbumForm(request.user, request.POST)
formset = PhotoFormSet(request.POST, request.FILES)
if albumform.is_valid() and formset.is_valid():
album = albumform.cleaned_data['album']
photo_ids = []
max_order = Photo.objects.filter(album=album).aggregate(Max('order'))['order__max'] or 0
order = max_order + 1
path = 'albums/%s/%s/' % (request.user.id, album.id)
# get the resizing dimensions from the preferences
preferences = Preferences.get_or_create(request.user)
resize_dimensions = preferences.get_default_img_size()
i = 0 # to access the file in request.FILES
for form in formset:
if form.has_changed():
instance = form.save(commit=False)
instance.user, instance.album = request.user, album
instance.order = order
img = request.FILES['form-%s-image' % i]
img_data = resize(img, upload_to=path, sizes_data=[resize_dimensions])
for data in img_data:
instance.width = data[1]
instance.height = data[2]
instance.image = data[0]
instance.save()
photo_ids.append(instance.id)
order += 1
i += 1
#make it a GET request again to pass it to the next function
request.method = "GET"
return set_extra_info(request, photo_ids, album)
else:
albumform = PickAlbumForm(request.user)
formset = PhotoFormSet(queryset=Photo.objects.none())
return render(request, 'albums/upload.html', {'amountform': amountform, 'albumform': albumform, 'formset': formset})
def descriptors_from_class(dataset, class_img_paths, class_number, option = constants.ORB_FEAT_OPTION):
"""
Gets all the local descriptors for a class. If an image has a side with more than 640 pixels it will be resized
leaving the biggest side at 640 pixels and conserving the aspect ratio for the other side.
Args:
dataset (Dataset object): An object that stores information about the dataset.
class_img_paths (array of strings): The paths for each image in certain class.
class_number (integer): The number of the class.
option (integer): If this is 49 (The key '1') uses ORB features, else use SIFT.
Returns:
numpy float matrix: Each row are the descriptors found in an image of the class
"""
des = None
step = (constants.STEP_PERCENTAGE * len(class_img_paths)) / 100
for i in range(len(class_img_paths)):
img_path = class_img_paths[i]
img = cv2.imread(img_path)
resize_to = 640
h, w, channels = img.shape
if h > resize_to or w > resize_to:
img = utils.resize(img, resize_to, h, w)
if option == constants.ORB_FEAT_OPTION:
des_name = "ORB"
new_des = orb(img)
else:
des_name = "SIFT"
new_des = sift(img)
if new_des is not None:
if des is None:
des = np.array(new_des, dtype=np.float32)
else:
des = np.vstack((des, np.array(new_des)))
# Print a message to show the status of the function
if i % step == 0:
percentage = (100 * i) / len(class_img_paths)
message = "Calculated {0} descriptors for image {1} of {2}({3}%) of class number {4} ...".format(
des_name, i, len(class_img_paths), percentage, class_number
)
print(message)
message = "* Finished getting the descriptors for the class number {0}*".format(class_number)
print(message)
print("Number of descriptors in class: {0}".format(len(des)))
dataset.set_class_count(class_number, len(des))
return des
def update_PCW(self, uNodeState, uInputInfo):
"""Update the PCW matrix."""
PCW = uNodeState['PCW']
k = uNodeState['k']
cls = uInputInfo['class']
try:
PCW[k, cls] += 1
except:
(rows, cols) = PCW.shape
(delta_r, delta_c) = (k + 1 - rows, cls + 1 - cols)
if delta_r < 0: delta_r = 0
if delta_c < 0: delta_c = 0
PCW = utils.resize(PCW, (rows + delta_r, cols + delta_c))
PCW[k, cls] = 1
uNodeState['PCW'] = PCW
def analyze(self):
"""
Character segmentation
"""
prev = 0
i = size
true_query = []
nimg = np.concatenate(
(np.zeros((20, size)), resize(self.img, width=self.img.shape[1], height=20), np.zeros((20, size))), axis=1
)
all_black = True
while i < self.img.shape[1]:
mimg = nimg[:, i - size : i + size]
mimg = mimg.reshape(-1).astype(float)
# nimg = pipe.transform(nimg)
probabilities = model.predict_proba(mimg)[0]
# print(probabilities)
if probabilities[0] > probabilities[1]:
pred = 0
else:
pred = 1
if all_black and nimg[:, i].sum() > 50:
all_black = False
# cv2.imwrite("tmp\\seg %d %s.jpg" % (i, pred), mimg.reshape((20,20)))
if pred == 1:
true_query.append((i - size, probabilities[1], all_black))
all_black = True
if 5 < i - prev < 30:
prev = i
i += 5
else:
prev = i
i += 1
self.segments = []
prev = 0
for i, prob, all_black in true_query:
if not all_black and nimg[:, prev + size : i + size].sum() > 200:
self.segments.append((prev, i))
prev = i
if nimg[:, prev + size :].sum() > 200:
self.segments.append((prev, self.img.shape[1] - 1))
def prepareByFile(imgObject, detector, w, h):
'''
resize, возвращение подготовленного для сравнения объекта(-ов)
:param imgObject: изображение
:param detector: путь до изображения
:param w: ширина изображения после resize
:param h: высота изображения после resize
:returns kp: особые точки изображения
:returns desc: описания особых точек изображения
'''
img = utils.readFromFile(imgObject)
img = utils.resize(img, w, h)
print 'PREPARE '
if not detector:
raise Exception("Detector can't be None")
kp, desc = detector.detectAndCompute(img, None)
return kp, desc
def prepare(imgPath, detector, w, h):
'''
Чтение изображения, resize, возвращение подготовленного для сравнения объекта(-ов)
:param imgPath: путь до изображения
:param detector: путь до изображения
:param w: ширина изображения после resize
:param h: высота изображения после resize
:returns kp: особые точки изображения
:returns desc: описания особых точек изображения
'''
img = utils.read(imgPath)
print img
if img == None:
raise Exception(u"Can't open file '%s'" % imgPath)
img = utils.resize(img, w, h)
print u'PREPARE %s' % imgPath
if not detector:
raise Exception("Detector can't be None")
kp, desc = detector.detectAndCompute(img, None)
return kp, desc
import utils
import constant
import glob
#temp, gain, bw, tau of power det
datafolder = constant.calibdatafolder+ '/2013_03_25/calib_board/'
files1 = glob.glob(datafolder+'C1pdboard200*')
files2 = glob.glob(datafolder+'C2pdboard200*')
mean1 = np.array([])
mean2 = np.array([])
vminusmean1 = np.array([])
vminusmean2 = np.array([])
for f1,f2 in zip(files1[::5],files2[::5]):
wf1 = utils.readscopefile(f1)
wf2 = utils.readscopefile(f2)
[pd,board] = utils.resize(wf1[1],wf2[1])
delay = utils.finddelay2(pd,board)
board = np.roll(board,delay)
m1 = np.mean(pd)
m2 = np.mean(board)
#DC component
mean1 = np.append(mean1,m1)
mean2 = np.append(mean2,m2)
#non DC component
vminusmean1 = np.append(vminusmean1, pd - m1)
vminusmean2 = np.append(vminusmean2, board - m2)
fig1 = plt.figure()
ax1 = plt.subplot(111)
ax1.plot(mean1,mean2,'ro',label='mean waveform')
fit = np.polyfit(mean1,mean2,1)
face_cascade = cv2.CascadeClassifier('haarcascade_frontalface_alt.xml')
body_cascade = cv2.CascadeClassifier('haarcascade_mcs_upperbody.xml')
if __name__ == '__main__':
arg = Arguments(sys.argv)
fullpath = arg.get_as_file(0)
# Check the file is exist
if fullpath is None:
sys.stderr.write(USAGE)
exit(1)
print('File:\t', fullpath)
img = cv2.imread(fullpath, cv2.IMREAD_COLOR)
img = resize(img, 300)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
faces = face_cascade.detectMultiScale(gray, 1.1, 3)
print('Face detect:\t', len(faces) > 0)
bodies = body_cascade.detectMultiScale(gray, 1.1, 3)
print('Body detect:\t', len(bodies) > 0)
if len(faces) <= 0 or len(bodies) <= 0:
exit(1)
face = Rectangle(faces[0][0], faces[0][1], faces[0][2], faces[0][3])
body = Rectangle(bodies[0][0], bodies[0][1], bodies[0][2], bodies[0][3])
# Skin Section
skinR = skinG = skinB = 0
#for f1, f2 in zip(filesRF, filesPD):
for f1, f2 in zip(filesRF[::5], filesPD[::5]):
mindist = 1e6
#result for each file
dist = np.array([])
besttau = 0
besta = 0
bestb = 0
wfRF = utils.readscopefile(f1)
wfPD = utils.readscopefile(f2)
for t in taus:
conv = utils.produceresponse(wfRF[0],wfRF[1],t)
real = wfPD[1]
sim = conv[1]
#resize the two waveforms to the same size (because of the convolution)
[real,sim] = utils.resize(real,sim)
time = utils.gettime(wfPD[0],conv[0])
delay = utils.finddelay2(real,sim)
simshifted = np.roll(sim,delay)
#fit the conv vs power:
fitconv_pd = np.polyfit(simshifted,real,1)
polyconv_pd = np.poly1d(fitconv_pd)
simpd = polyconv_pd(simshifted)
size = len(simpd)
alpha = np.sum( (simpd - real)**2)
dist = np.append(dist,alpha)
if alpha < mindist:
mindist = alpha
besttau = t
besta = fitconv_pd[0]
bestb = fitconv_pd[1]
X = []
y = []
resize_to = 640
#des = None
print ('start loading ' + str(len(pos_img_files)-1) + ' positive files')
for pos_img_file in pos_img_files:
pos_filepath = pos_img_dir + pos_img_file
# not include hidden files
root, ext = os.path.splitext(pos_filepath)
if ext == ".JPG" :
print(pos_filepath)
pos_img = cv2.imread(pos_filepath)
h, w, channels = pos_img.shape
if h > resize_to or w > resize_to:
pos_img = utils.resize(pos_img, resize_to, h, w)
gray_pos= cv2.cvtColor(pos_img,cv2.COLOR_BGR2GRAY)
#orb
orb = cv2.ORB_create()
kp_orb, des = orb.detectAndCompute(gray_pos, None)
img_orb = cv2.drawKeypoints(gray_pos, kp_orb, None)
#akaze
#akaze = cv2.AKAZE_create()
#kp_akaze, des = akaze.detectAndCompute(gray_pos, None)
#img_akaze = cv2.drawKeypoints(gray_pos, kp_akaze, None)
#fast
#detect keypoints
#fast = cv2.FastFeatureDetector_create()
def __init__(self, begin, end, img):
self.img = resize(img, height=Line.line_height)
self.beginY = begin
self.endY = end