def main(args):
(image, cnts, pixelsPerMetric) = mark_countor(args['image'])
for cnt in cnts:
# ignore the contour if the area is small(noise)
if cv2.contourArea(cnt) < Config.AREA_THRESHOLD:
continue
visual(image, cnt, pixelsPerMetric, args['width'])
def text_detect_CRAFT(img,
craft_config,
CRAFT_MODEL,
sortbb=True,
visual_img=False):
'''
args:
img: image
craft_config: config of craft
CRAFT_MODEL: craft model
sort_bb: whether or not sort bounding box
visual_image: whether or no not visual image
return:
bboxes: bbox of text
polys: polygon of text
score_text: confidence score
'''
# img = loadImage(image_path)
bboxes, polys, score_text = craft_text_detect(img, craft_config,
CRAFT_MODEL)
if sortbb:
bboxes = sort_bb(bboxes)
if visual_img:
img = visual(img, polys)
return bboxes, polys, score_text
def visualization(sess, model, data_loader, filename):
saver = tf.train.Saver(max_to_keep=1)
ckpt = tf.train.get_checkpoint_state(flags.ckpt_dir)
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
print(" [*] loading parameters success !!!")
else:
print(" [!] loading parameters failed ...")
return
# user,item,label, utexts, itexts, text= data_loader.sample_point()
user, item, label, utexts, itexts, text = data_loader.find_a_user()
utexts = utexts.astype(int)
itexts = itexts.astype(int)
feed_dict = {
model.u_input: user,
model.i_input: item,
model.label: label,
model.utext: utexts,
model.itext: itexts,
model.text: text,
model.keep_prob: 1.0
}
res = sess.run([
model.word_user_alpha, model.word_item_alpha, model.doc_user_alpha,
model.doc_item_alpha
],
feed_dict=feed_dict)
print(utexts.dtype)
u_texts = data_loader.vec_texts[utexts]
i_texts = data_loader.vec_texts[itexts]
res[2] = np.array(res[2]).transpose(1, 0, 2)
res[3] = np.array(res[3]).transpose(1, 0, 2)
for i in range(len(user)):
uit = [user[i], item[i], label[i]]
print(uit)
res_trans = []
for r in res:
res_trans.append(r[i])
visual(res_trans, uit, data_loader, utexts[i], itexts[i], u_texts[i],
i_texts[i], filename)
def cluster_image(img, colorspace, clusters, func, name):
if colorspace is not None:
img = cv2.cvtColor(img, colorspace)
points = utils.image_to_points(img)
labels = func(points, clusters)
res, mask = utils.visual(points, labels, img.shape, name, colorspace)
return res, labels, mask
def text_detect_CRAFT(image_path,
craft_config,
net_craft,
sortbb=True,
visual_img=True):
img = loadImage(image_path)
bboxes, polys, score_text = craft_text_detect(img, craft_config, net_craft)
if sortbb:
polys = sorting_bounding_box(polys)
if visual_img:
img = visual(img, polys)
return img, bboxes, polys, score_text
Download weigth của craft trên drive về bỏ vào thư mục libs/CRAFT/models
Chạy file này, thay link ảnh đầu vào, đầu ra của file này sẽ trả về danh sách xmin, ymin, xmax, ymax của các bounding box
đã được sắp xếp theo thứ tự từ trái qua phải, từ trên xuống dưới
'''
def loadImage(img_file):
img = cv2.imread(img_file) # RGB order
if img.shape[0] == 2: img = img[0]
if len(img.shape) == 2: img = cv2.cvtColor(img, cv2.COLOR_GRAY2RGB)
if img.shape[2] == 4: img = img[:, :, :3]
img = np.array(img)
return img
# setup config
cfg = get_config()
cfg.merge_from_file('configs/craft.yaml')
craft_config = cfg.CRAFT
# run craft
net = CRAFT()
img = loadImage('data/test.jpg')
print('--------craft processing----------')
bboxes, polys, score_text = craft_text_detect(img, craft_config, net)
polys = sorting_bounding_box(polys)
# hàm lưu ảnh kết quả xem chơi
visual(img, polys)
print('--------craft done ----------')
def __call__(self, image, targets, input_dim):
boxes = targets[:, :4].copy()
labels = targets[:, 4].copy()
if targets.shape[1] > 5:
mixup = True
ratios = targets[:, -1].copy()
ratios_o = targets[:, -1].copy()
else:
mixup = False
ratios = None
ratios_o = None
lshape = 6 if mixup else 5
if len(boxes) == 0:
targets = np.zeros((self.max_labels, lshape), dtype=np.float32)
image = preproc_for_test(image, input_dim, self.means, self.std)
image = np.ascontiguousarray(image, dtype=np.float32)
return torch.from_numpy(image), torch.from_numpy(targets)
image_o = image.copy()
targets_o = targets.copy()
height_o, width_o, _ = image_o.shape
boxes_o = targets_o[:, :4]
labels_o = targets_o[:, 4]
b_x_o = (boxes_o[:, 2] + boxes_o[:, 0]) * 0.5
b_y_o = (boxes_o[:, 3] + boxes_o[:, 1]) * 0.5
b_w_o = (boxes_o[:, 2] - boxes_o[:, 0]) * 1.0
b_h_o = (boxes_o[:, 3] - boxes_o[:, 1]) * 1.0
boxes_o[:, 0] = b_x_o
boxes_o[:, 1] = b_y_o
boxes_o[:, 2] = b_w_o
boxes_o[:, 3] = b_h_o
boxes_o[:, 0::2] /= width_o
boxes_o[:, 1::2] /= height_o
boxes_o[:, 0::2] *= input_dim[0]
boxes_o[:, 1::2] *= input_dim[1]
# labels_o = np.expand_dims(labels_o,1)
# targets_o = np.hstack((boxes_o,labels_o))
# targets_o = np.hstack((labels_o,boxes_o))
image_t = _distort(image)
if self.means is not None:
fill = [m * 255 for m in self.means]
fill = fill[::-1]
else:
fill = (127.5, 127.5, 127.5)
image_t, boxes = _expand(image_t, boxes, fill, self.p)
image_t, boxes, labels, ratios = _crop(image_t, boxes, labels, ratios)
image_t, boxes = _mirror(image_t, boxes)
if random.randrange(2):
image_t, boxes, _ = _random_affine(image_t,
boxes,
borderValue=fill)
height, width, _ = image_t.shape
if DEBUG:
image_t = np.ascontiguousarray(image_t, dtype=np.uint8)
img = visual(image_t, boxes, labels)
cv2.imshow("DEBUG", img)
cv2.waitKey(0)
image_t = preproc_for_test(image_t, input_dim, self.means, self.std)
boxes = boxes.copy()
b_x = (boxes[:, 2] + boxes[:, 0]) * 0.5
b_y = (boxes[:, 3] + boxes[:, 1]) * 0.5
b_w = (boxes[:, 2] - boxes[:, 0]) * 1.0
b_h = (boxes[:, 3] - boxes[:, 1]) * 1.0
boxes[:, 0] = b_x
boxes[:, 1] = b_y
boxes[:, 2] = b_w
boxes[:, 3] = b_h
boxes[:, 0::2] /= width
boxes[:, 1::2] /= height
boxes[:, 0::2] *= input_dim[0]
boxes[:, 1::2] *= input_dim[1]
mask_b = np.minimum(boxes[:, 2], boxes[:, 3]) > 6
# mask_b= (boxes[:,2]*boxes[:,3]) > 32**2
# mask_b= (boxes[:,2]*boxes[:,3]) > 48**2
boxes_t = boxes[mask_b]
labels_t = labels[mask_b].copy()
if mixup:
ratios_t = ratios[mask_b].copy()
"""
if len(boxes_t)==0:
targets = np.zeros((self.max_labels,lshape),dtype=np.float32)
image = preproc_for_test(image_o, input_dim, self.means, self.std)
image = np.ascontiguousarray(image, dtype=np.float32)
return torch.from_numpy(image), torch.from_numpy(targets)
"""
# if len(boxes_t)==0 or random.random() > 0.97:
if len(boxes_t) == 0:
image_t = preproc_for_test(image_o, input_dim, self.means,
self.std)
boxes_t = boxes_o
labels_t = labels_o
ratios_t = ratios_o
labels_t = np.expand_dims(labels_t, 1)
if mixup:
ratios_t = np.expand_dims(ratios_t, 1)
targets_t = np.hstack((labels_t, boxes_t, ratios_t))
else:
targets_t = np.hstack((labels_t, boxes_t))
padded_labels = np.zeros((self.max_labels, lshape))
padded_labels[range(
len(targets_t))[:self.max_labels]] = targets_t[:self.max_labels]
padded_labels = np.ascontiguousarray(padded_labels, dtype=np.float32)
image_t = np.ascontiguousarray(image_t, dtype=np.float32)
return torch.from_numpy(image_t), torch.from_numpy(padded_labels)