def fake_char_boxes(net, src, word_box, word_length):
img, src_points, crop_points = crop_image(src, word_box, dst_height=64.)
h, w = img.shape[:2]
if min(h, w) == 0:
confidence = 0.5
region_boxes = divide_region(word_box, word_length)
region_boxes = [
reorder_points(region_box) for region_box in region_boxes
]
return region_boxes, confidence
img = img_normalize(img)
# print(img.shape)
region_score, _ = net.predict(np.array([img]))
heat_map = region_score[0] * 255.
heat_map = heat_map.astype(np.uint8)
marker_map = watershed(heat_map)
region_boxes = find_box(marker_map)
confidence = cal_confidence(region_boxes, word_length)
if confidence <= 0.5:
confidence = 0.5
region_boxes = divide_region(word_box, word_length)
region_boxes = [
reorder_points(region_box) for region_box in region_boxes
]
else:
region_boxes = np.array(region_boxes) * 2
region_boxes = enlarge_char_boxes(region_boxes, crop_points)
region_boxes = [
un_warping(region_box, src_points, crop_points)
for region_box in region_boxes
]
# print(word_box, region_boxes)
return region_boxes, confidence
def fake_char_boxes(self, src, word_box, word_length):
img, src_points, crop_points = crop_image(src,
word_box,
dst_height=64.)
h, w = img.shape[:2]
if min(h, w) == 0:
confidence = 0.5
region_boxes = divide_region(word_box, word_length)
region_boxes = [
reorder_points(region_box) for region_box in region_boxes
]
return region_boxes, confidence
img = img_normalize(img)
region_score, affinity_score = self.test_net(self.craft, img,
self.cuda)
heat_map = region_score * 255.
heat_map = heat_map.astype(np.uint8)
marker_map = watershed(heat_map)
region_boxes = find_box(marker_map)
confidence = cal_confidence(region_boxes, word_length)
if confidence <= 0.5:
confidence = 0.5
region_boxes = divide_region(word_box, word_length)
region_boxes = [
reorder_points(region_box) for region_box in region_boxes
]
else:
region_boxes = divide_region(word_box, word_length)
region_boxes = [
reorder_points(region_box) for region_box in region_boxes
]
return region_boxes, confidence
def test_net(net, image, text_threshold, link_threshold, low_text, poly,
refine_net):
# t0 = time.time()
# resize
img_resized, target_ratio, size_heatmap = resize_aspect_ratio(
image,
args.canvas_size,
interpolation=cv2.INTER_LINEAR,
mag_ratio=args.mag_ratio)
ratio_h = ratio_w = 1 / target_ratio
# preprocessing
x = img_normalize(img_resized)
x = np.expand_dims(x, axis=0) # [h, w, c] to [b, h, w, c]
# forward pass
y, feature = net(x)
# make score and link map
score_text = y[0, :, :, 0]
score_link = y[0, :, :, 1]
# refine link
if refine_net is not None:
# TODO
pass
# t0 = time.time() - t0
# t1 = time.time()
# Post-processing
boxes, polys = getDetBoxes(score_text, score_link, text_threshold,
link_threshold, low_text, poly)
# coordinate adjustment
boxes = adjustResultCoordinates(boxes, ratio_w, ratio_h)
polys = adjustResultCoordinates(polys, ratio_w, ratio_h)
for k in range(len(polys)):
if polys[k] is None:
polys[k] = boxes[k]
# t1 = time.time() - t1
# render results (optional)
render_img = score_text.numpy()
render_img = np.hstack((render_img, np.ones(
(np.shape(render_img)[0], 5)), score_link))
ret_score_text = score_to_heat_map(render_img)
# if args.show_time : print("\ninfer/postproc time : {:.3f}/{:.3f}".format(t0, t1))
return boxes, polys, ret_score_text
def predict(model, image, text_threshold, link_threshold, low_text):
t0 = time.time()
# resize
h, w = image.shape[:2]
mag_ratio = 600 / max(h, w)
# img_resized, target_ratio = img_resize(image, FLAGS.mag_ratio, FLAGS.canvas_size, interpolation=cv2.INTER_LINEAR)
img_resized, target_ratio = img_resize(image,
mag_ratio,
FLAGS.canvas_size,
interpolation=cv2.INTER_LINEAR)
ratio_h = ratio_w = 1 / target_ratio
# preprocessing
x = img_normalize(img_resized)
# make score and link map
score_text, score_link = model.predict(np.array([x]))
score_text = score_text[0]
score_link = score_link[0]
t0 = time.time() - t0
t1 = time.time()
# Post-processing
boxes = getDetBoxes(score_text, score_link, text_threshold, link_threshold,
low_text)
boxes = adjustResultCoordinates(boxes, ratio_w, ratio_h)
t1 = time.time() - t1
# render results (optional)
render_img = score_text.copy()
white_img = np.ones((render_img.shape[0], 10, 3), dtype=np.uint8) * 255
ret_score_text = np.hstack(
(to_heat_map(render_img), white_img, to_heat_map(score_link)))
# ret_score_text = to_heat_map(render_img)
if FLAGS.show_time:
print("\ninfer/postproc time : {:.3f}/{:.3f}".format(t0, t1))
return boxes, ret_score_text