class OCRService(WebService):
def init_rec(self):
self.ocr_reader = OCRReader()
def preprocess(self, feed=[], fetch=[]):
img_list = []
for feed_data in feed:
data = base64.b64decode(feed_data["image"].encode('utf8'))
data = np.fromstring(data, np.uint8)
im = cv2.imdecode(data, cv2.IMREAD_COLOR)
img_list.append(im)
max_wh_ratio = 0
for i, boximg in enumerate(img_list):
h, w = boximg.shape[0:2]
wh_ratio = w * 1.0 / h
max_wh_ratio = max(max_wh_ratio, wh_ratio)
_, w, h = self.ocr_reader.resize_norm_img(img_list[0],
max_wh_ratio).shape
imgs = np.zeros((len(img_list), 3, w, h)).astype('float32')
for i, img in enumerate(img_list):
norm_img = self.ocr_reader.resize_norm_img(img, max_wh_ratio)
imgs[i] = norm_img
feed = {"image": imgs.copy()}
fetch = ["ctc_greedy_decoder_0.tmp_0", "softmax_0.tmp_0"]
return feed, fetch
def postprocess(self, feed={}, fetch=[], fetch_map=None):
rec_res = self.ocr_reader.postprocess(fetch_map, with_score=True)
res_lst = []
for res in rec_res:
res_lst.append(res[0])
res = {"res": res_lst}
return res
class OCRService(WebService):
def init_det_client(self, det_port, det_client_config):
self.det_preprocess = Sequential([
ResizeByFactor(32, 960),
Div(255),
Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
Transpose((2, 0, 1))
])
self.det_client = Client()
self.det_client.load_client_config(det_client_config)
self.det_client.connect(["127.0.0.1:{}".format(det_port)])
self.ocr_reader = OCRReader()
def preprocess(self, feed=[], fetch=[]):
data = base64.b64decode(feed[0]["image"].encode('utf8'))
data = np.fromstring(data, np.uint8)
im = cv2.imdecode(data, cv2.IMREAD_COLOR)
ori_h, ori_w, _ = im.shape
det_img = self.det_preprocess(im)
det_out = self.det_client.predict(feed={"image": det_img},
fetch=["concat_1.tmp_0"],
batch=False)
_, new_h, new_w = det_img.shape
filter_func = FilterBoxes(10, 10)
post_func = DBPostProcess({
"thresh": 0.3,
"box_thresh": 0.5,
"max_candidates": 1000,
"unclip_ratio": 1.5,
"min_size": 3
})
sorted_boxes = SortedBoxes()
ratio_list = [float(new_h) / ori_h, float(new_w) / ori_w]
dt_boxes_list = post_func(det_out["concat_1.tmp_0"], [ratio_list])
dt_boxes = filter_func(dt_boxes_list[0], [ori_h, ori_w])
dt_boxes = sorted_boxes(dt_boxes)
get_rotate_crop_image = GetRotateCropImage()
feed_list = []
img_list = []
max_wh_ratio = 0
for i, dtbox in enumerate(dt_boxes):
boximg = get_rotate_crop_image(im, dt_boxes[i])
img_list.append(boximg)
h, w = boximg.shape[0:2]
wh_ratio = w * 1.0 / h
max_wh_ratio = max(max_wh_ratio, wh_ratio)
for img in img_list:
norm_img = self.ocr_reader.resize_norm_img(img, max_wh_ratio)
feed_list.append(norm_img[np.newaxis, :])
feed_batch = {"image": np.concatenate(feed_list, axis=0)}
fetch = ["ctc_greedy_decoder_0.tmp_0", "softmax_0.tmp_0"]
return feed_batch, fetch, True
def postprocess(self, feed={}, fetch=[], fetch_map=None):
rec_res = self.ocr_reader.postprocess(fetch_map, with_score=True)
res_lst = []
for res in rec_res:
res_lst.append(res[0])
res = {"res": res_lst}
return res
class OCRService(WebService):
def init_rec(self):
self.ocr_reader = OCRReader()
def preprocess(self, feed=[], fetch=[]):
# TODO: to handle batch rec images
img_list = []
for feed_data in feed:
data = base64.b64decode(feed_data["image"].encode('utf8'))
data = np.fromstring(data, np.uint8)
im = cv2.imdecode(data, cv2.IMREAD_COLOR)
img_list.append(im)
feed_list = []
max_wh_ratio = 0
for i, boximg in enumerate(img_list):
h, w = boximg.shape[0:2]
wh_ratio = w * 1.0 / h
max_wh_ratio = max(max_wh_ratio, wh_ratio)
for img in img_list:
norm_img = self.ocr_reader.resize_norm_img(img, max_wh_ratio)
feed = {"image": norm_img}
feed_list.append(feed)
fetch = ["ctc_greedy_decoder_0.tmp_0", "softmax_0.tmp_0"]
return feed_list, fetch
def postprocess(self, feed={}, fetch=[], fetch_map=None):
rec_res = self.ocr_reader.postprocess(fetch_map, with_score=True)
res_lst = []
for res in rec_res:
res_lst.append(res[0])
res = {"res": res_lst}
return res
class OCRService(WebService):
def init_det_debugger(self, det_model_config):
self.det_preprocess = Sequential([
ResizeByFactor(32, 960),
Div(255),
Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
Transpose((2, 0, 1))
])
self.det_client = Debugger()
if sys.argv[1] == 'gpu':
self.det_client.load_model_config(det_model_config,
gpu=True,
profile=False)
elif sys.argv[1] == 'cpu':
self.det_client.load_model_config(det_model_config,
gpu=False,
profile=False)
self.ocr_reader = OCRReader()
def preprocess(self, feed=[], fetch=[]):
data = base64.b64decode(feed[0]["image"].encode('utf8'))
data = np.fromstring(data, np.uint8)
im = cv2.imdecode(data, cv2.IMREAD_COLOR)
ori_h, ori_w, _ = im.shape
det_img = self.det_preprocess(im)
_, new_h, new_w = det_img.shape
det_img = det_img[np.newaxis, :]
det_img = det_img.copy()
det_out = self.det_client.predict(feed={"image": det_img},
fetch=["concat_1.tmp_0"])
filter_func = FilterBoxes(10, 10)
post_func = DBPostProcess({
"thresh": 0.3,
"box_thresh": 0.5,
"max_candidates": 1000,
"unclip_ratio": 1.5,
"min_size": 3
})
sorted_boxes = SortedBoxes()
ratio_list = [float(new_h) / ori_h, float(new_w) / ori_w]
dt_boxes_list = post_func(det_out["concat_1.tmp_0"], [ratio_list])
dt_boxes = filter_func(dt_boxes_list[0], [ori_h, ori_w])
dt_boxes = sorted_boxes(dt_boxes)
get_rotate_crop_image = GetRotateCropImage()
img_list = []
max_wh_ratio = 0
for i, dtbox in enumerate(dt_boxes):
boximg = get_rotate_crop_image(im, dt_boxes[i])
img_list.append(boximg)
h, w = boximg.shape[0:2]
wh_ratio = w * 1.0 / h
max_wh_ratio = max(max_wh_ratio, wh_ratio)
if len(img_list) == 0:
return [], []
_, w, h = self.ocr_reader.resize_norm_img(img_list[0],
max_wh_ratio).shape
imgs = np.zeros((len(img_list), 3, w, h)).astype('float32')
for id, img in enumerate(img_list):
norm_img = self.ocr_reader.resize_norm_img(img, max_wh_ratio)
imgs[id] = norm_img
feed = {"image": imgs.copy()}
fetch = ["ctc_greedy_decoder_0.tmp_0", "softmax_0.tmp_0"]
return feed, fetch
def postprocess(self, feed={}, fetch=[], fetch_map=None):
rec_res = self.ocr_reader.postprocess(fetch_map, with_score=True)
res_lst = []
for res in rec_res:
res_lst.append(res[0])
res = {"res": res_lst}
return res