Skip to content

P
PaddleOCR

s920243400 / PaddleOCR
与 Fork 源项目一致

Fork自 PaddlePaddle / PaddleOCR

通知 1
Star 1
Fork 0
P
PaddleOCR
提交 3f5ffc32 编写于 作者: T tink2123
浏览文件
操作

add paddle2onnx doc

上级 1c5d1aa4
隐藏空白更改
内联 并排
Showing with 317 addition and 1281 deletion
deploy/paddle2onnx/predict_cls.py 已删除 100644 → 0
浏览文件 @ 1c5d1aa4
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import os
import sys
__dir__ = os.path.dirname(os.path.abspath(__file__))
sys.path.append(__dir__)
sys.path.append(os.path.abspath(os.path.join(__dir__, '../..')))
os.environ["FLAGS_allocator_strategy"] = 'auto_growth'
import cv2
import copy
import numpy as np
import math
import time
import traceback
import utility
from ppocr.postprocess import build_post_process
from ppocr.utils.logging import get_logger
from ppocr.utils.utility import get_image_file_list, check_and_read_gif
logger = get_logger()
class TextClassifier(object):
def __init__(self, args):
self.cls_image_shape = [int(v) for v in args.cls_image_shape.split(",")]
self.cls_batch_num = args.cls_batch_num
self.cls_thresh = args.cls_thresh
postprocess_params = {
'name': 'ClsPostProcess',
"label_list": args.label_list,
}
self.postprocess_op = build_post_process(postprocess_params)
self.predictor, self.input_tensor, self.output_tensors = \
utility.create_predictor(args, 'cls', logger)
def resize_norm_img(self, img):
imgC, imgH, imgW = self.cls_image_shape
h = img.shape[0]
w = img.shape[1]
ratio = w / float(h)
if math.ceil(imgH * ratio) > imgW:
resized_w = imgW
else:
resized_w = int(math.ceil(imgH * ratio))
resized_image = cv2.resize(img, (resized_w, imgH))
resized_image = resized_image.astype('float32')
if self.cls_image_shape[0] == 1:
resized_image = resized_image / 255
resized_image = resized_image[np.newaxis, :]
else:
resized_image = resized_image.transpose((2, 0, 1)) / 255
resized_image -= 0.5
resized_image /= 0.5
padding_im = np.zeros((imgC, imgH, imgW), dtype=np.float32)
padding_im[:, :, 0:resized_w] = resized_image
return padding_im
def __call__(self, img_list):
img_list = copy.deepcopy(img_list)
img_num = len(img_list)
# Calculate the aspect ratio of all text bars
width_list = []
for img in img_list:
width_list.append(img.shape[1] / float(img.shape[0]))
# Sorting can speed up the cls process
indices = np.argsort(np.array(width_list))
cls_res = [['', 0.0]] * img_num
batch_num = self.cls_batch_num
elapse = 0
for beg_img_no in range(0, img_num, batch_num):
end_img_no = min(img_num, beg_img_no + batch_num)
norm_img_batch = []
max_wh_ratio = 0
for ino in range(beg_img_no, end_img_no):
h, w = img_list[indices[ino]].shape[0:2]
wh_ratio = w * 1.0 / h
max_wh_ratio = max(max_wh_ratio, wh_ratio)
for ino in range(beg_img_no, end_img_no):
norm_img = self.resize_norm_img(img_list[indices[ino]])
norm_img = norm_img[np.newaxis, :]
norm_img_batch.append(norm_img)
norm_img_batch = np.concatenate(norm_img_batch)
norm_img_batch = norm_img_batch.copy()
starttime = time.time()
input_dict = {}
input_dict[self.input_tensor.name] = norm_img_batch
outputs = self.predictor.run(self.output_tensors, input_dict)
prob_out = outputs[0]
cls_result = self.postprocess_op(prob_out)
elapse += time.time() - starttime
for rno in range(len(cls_result)):
label, score = cls_result[rno]
cls_res[indices[beg_img_no + rno]] = [label, score]
if '180' in label and score > self.cls_thresh:
img_list[indices[beg_img_no + rno]] = cv2.rotate(
img_list[indices[beg_img_no + rno]], 1)
return img_list, cls_res, elapse
def main(args):
image_file_list = get_image_file_list(args.image_dir)
text_classifier = TextClassifier(args)
valid_image_file_list = []
img_list = []
for image_file in image_file_list:
img, flag = check_and_read_gif(image_file)
if not flag:
img = cv2.imread(image_file)
if img is None:
logger.info("error in loading image:{}".format(image_file))
continue
valid_image_file_list.append(image_file)
img_list.append(img)
try:
img_list, cls_res, predict_time = text_classifier(img_list)
except:
logger.info(traceback.format_exc())
logger.info(
"ERROR!!!! \n"
"Please read the FAQ：https://github.com/PaddlePaddle/PaddleOCR#faq \n"
"If your model has tps module: "
"TPS does not support variable shape.\n"
"Please set --rec_image_shape='3,32,100' and --rec_char_type='en' ")
exit()
for ino in range(len(img_list)):
logger.info("Predicts of {}:{}".format(valid_image_file_list[ino],
cls_res[ino]))
logger.info("Total predict time for {} images, cost: {:.3f}".format(
len(img_list), predict_time))
if __name__ == "__main__":
main(utility.parse_args())
deploy/paddle2onnx/predict_det.py 已删除 100644 → 0
浏览文件 @ 1c5d1aa4
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import os
import sys
__dir__ = os.path.dirname(os.path.abspath(__file__))
sys.path.append(__dir__)
sys.path.append(os.path.abspath(os.path.join(__dir__, '../..')))
os.environ["FLAGS_allocator_strategy"] = 'auto_growth'
import cv2
import numpy as np
import time
import sys
import utility as utility
from ppocr.utils.logging import get_logger
from ppocr.utils.utility import get_image_file_list, check_and_read_gif
from ppocr.data import create_operators, transform
from ppocr.postprocess import build_post_process
logger = get_logger()
class TextDetector(object):
def __init__(self, args):
self.args = args
self.det_algorithm = args.det_algorithm
pre_process_list = [{
'DetResizeForTest': {
'image_shape': [640, 640]
}
}, {
'NormalizeImage': {
'std': [0.229, 0.224, 0.225],
'mean': [0.485, 0.456, 0.406],
'scale': '1./255.',
'order': 'hwc'
}
}, {
'ToCHWImage': None
}, {
'KeepKeys': {
'keep_keys': ['image', 'shape']
}
}]
postprocess_params = {}
if self.det_algorithm == "DB":
postprocess_params['name'] = 'DBPostProcess'
postprocess_params["thresh"] = args.det_db_thresh
postprocess_params["box_thresh"] = args.det_db_box_thresh
postprocess_params["max_candidates"] = 1000
postprocess_params["unclip_ratio"] = args.det_db_unclip_ratio
postprocess_params["use_dilation"] = True
elif self.det_algorithm == "EAST":
postprocess_params['name'] = 'EASTPostProcess'
postprocess_params["score_thresh"] = args.det_east_score_thresh
postprocess_params["cover_thresh"] = args.det_east_cover_thresh
postprocess_params["nms_thresh"] = args.det_east_nms_thresh
elif self.det_algorithm == "SAST":
pre_process_list[0] = {
'DetResizeForTest': {
'resize_long': args.det_limit_side_len
}
}
postprocess_params['name'] = 'SASTPostProcess'
postprocess_params["score_thresh"] = args.det_sast_score_thresh
postprocess_params["nms_thresh"] = args.det_sast_nms_thresh
self.det_sast_polygon = args.det_sast_polygon
if self.det_sast_polygon:
postprocess_params["sample_pts_num"] = 6
postprocess_params["expand_scale"] = 1.2
postprocess_params["shrink_ratio_of_width"] = 0.2
else:
postprocess_params["sample_pts_num"] = 2
postprocess_params["expand_scale"] = 1.0
postprocess_params["shrink_ratio_of_width"] = 0.3
else:
logger.info("unknown det_algorithm:{}".format(self.det_algorithm))
sys.exit(0)
self.preprocess_op = create_operators(pre_process_list)
self.postprocess_op = build_post_process(postprocess_params)
print()
self.predictor, self.input_tensor, self.output_tensors = utility.create_predictor(
args, 'det', logger) # paddle.jit.load(args.det_model_dir)
# self.predictor.eval()
def order_points_clockwise(self, pts):
"""
reference from: https://github.com/jrosebr1/imutils/blob/master/imutils/perspective.py
# sort the points based on their x-coordinates
"""
xSorted = pts[np.argsort(pts[:, 0]), :]
# grab the left-most and right-most points from the sorted
# x-roodinate points
leftMost = xSorted[:2, :]
rightMost = xSorted[2:, :]
# now, sort the left-most coordinates according to their
# y-coordinates so we can grab the top-left and bottom-left
# points, respectively
leftMost = leftMost[np.argsort(leftMost[:, 1]), :]
(tl, bl) = leftMost
rightMost = rightMost[np.argsort(rightMost[:, 1]), :]
(tr, br) = rightMost
rect = np.array([tl, tr, br, bl], dtype="float32")
return rect
def clip_det_res(self, points, img_height, img_width):
for pno in range(points.shape[0]):
points[pno, 0] = int(min(max(points[pno, 0], 0), img_width - 1))
points[pno, 1] = int(min(max(points[pno, 1], 0), img_height - 1))
return points
def filter_tag_det_res(self, dt_boxes, image_shape):
img_height, img_width = image_shape[0:2]
dt_boxes_new = []
for box in dt_boxes:
box = self.order_points_clockwise(box)
box = self.clip_det_res(box, img_height, img_width)
rect_width = int(np.linalg.norm(box[0] - box[1]))
rect_height = int(np.linalg.norm(box[0] - box[3]))
if rect_width <= 3 or rect_height <= 3:
continue
dt_boxes_new.append(box)
dt_boxes = np.array(dt_boxes_new)
return dt_boxes
def filter_tag_det_res_only_clip(self, dt_boxes, image_shape):
img_height, img_width = image_shape[0:2]
dt_boxes_new = []
for box in dt_boxes:
box = self.clip_det_res(box, img_height, img_width)
dt_boxes_new.append(box)
dt_boxes = np.array(dt_boxes_new)
return dt_boxes
def __call__(self, img):
ori_im = img.copy()
data = {'image': img}
data = transform(data, self.preprocess_op)
img, shape_list = data
if img is None:
return None, 0
img = np.expand_dims(img, axis=0)
shape_list = np.expand_dims(shape_list, axis=0)
img = img.copy()
starttime = time.time()
input_dict = {}
input_dict[self.input_tensor.name] = img
outputs = self.predictor.run(self.output_tensors, input_dict)
preds = {}
if self.det_algorithm == "EAST":
preds['f_geo'] = outputs[0]
preds['f_score'] = outputs[1]
elif self.det_algorithm == 'SAST':
preds['f_border'] = outputs[0]
preds['f_score'] = outputs[1]
preds['f_tco'] = outputs[2]
preds['f_tvo'] = outputs[3]
elif self.det_algorithm == 'DB':
preds['maps'] = outputs[0]
else:
raise NotImplementedError
post_result = self.postprocess_op(preds, shape_list)
dt_boxes = post_result[0]['points']
if self.det_algorithm == "SAST" and self.det_sast_polygon:
dt_boxes = self.filter_tag_det_res_only_clip(dt_boxes, ori_im.shape)
else:
dt_boxes = self.filter_tag_det_res(dt_boxes, ori_im.shape)
elapse = time.time() - starttime
return dt_boxes, elapse
if __name__ == "__main__":
args = utility.parse_args()
image_file_list = get_image_file_list(args.image_dir)
text_detector = TextDetector(args)
count = 0
total_time = 0
draw_img_save = "./inference_results"
if not os.path.exists(draw_img_save):
os.makedirs(draw_img_save)
for image_file in image_file_list:
img, flag = check_and_read_gif(image_file)
if not flag:
img = cv2.imread(image_file)
if img is None:
logger.info("error in loading image:{}".format(image_file))
continue
dt_boxes, elapse = text_detector(img)
if count > 0:
total_time += elapse
count += 1
logger.info("Predict time of {}: {}".format(image_file, elapse))
src_im = utility.draw_text_det_res(dt_boxes, image_file)
img_name_pure = os.path.split(image_file)[-1]
img_path = os.path.join(draw_img_save,
"det_res_{}".format(img_name_pure))
cv2.imwrite(img_path, src_im)
logger.info("The visualized image saved in {}".format(img_path))
if count > 1:
logger.info("Avg Time: {}".format(total_time / (count - 1)))
deploy/paddle2onnx/predict_rec.py 已删除 100644 → 0
浏览文件 @ 1c5d1aa4
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import os
import sys
__dir__ = os.path.dirname(os.path.abspath(__file__))
sys.path.append(__dir__)
sys.path.append(os.path.abspath(os.path.join(__dir__, '../..')))
os.environ["FLAGS_allocator_strategy"] = 'auto_growth'
import cv2
import numpy as np
import math
import time
import traceback
import utility
from ppocr.postprocess import build_post_process
from ppocr.utils.logging import get_logger
from ppocr.utils.utility import get_image_file_list, check_and_read_gif
logger = get_logger()
class TextRecognizer(object):
def __init__(self, args):
self.rec_image_shape = [int(v) for v in args.rec_image_shape.split(",")]
self.character_type = args.rec_char_type
self.rec_batch_num = args.rec_batch_num
self.rec_algorithm = args.rec_algorithm
postprocess_params = {
'name': 'CTCLabelDecode',
"character_type": args.rec_char_type,
"character_dict_path": args.rec_char_dict_path,
"use_space_char": args.use_space_char
}
self.postprocess_op = build_post_process(postprocess_params)
self.predictor, self.input_tensor, self.output_tensors = \
utility.create_predictor(args, 'rec', logger)
def resize_norm_img(self, img, max_wh_ratio):
imgC, imgH, imgW = self.rec_image_shape
assert imgC == img.shape[2]
# if self.character_type == "ch":
# imgW = int((32 * max_wh_ratio))
h, w = img.shape[:2]
ratio = w / float(h)
if math.ceil(imgH * ratio) > imgW:
resized_w = imgW
else:
resized_w = int(math.ceil(imgH * ratio))
resized_image = cv2.resize(img, (resized_w, imgH))
resized_image = resized_image.astype('float32')
resized_image = resized_image.transpose((2, 0, 1)) / 255
resized_image -= 0.5
resized_image /= 0.5
padding_im = np.zeros((imgC, imgH, imgW), dtype=np.float32)
padding_im[:, :, 0:resized_w] = resized_image
return padding_im
def __call__(self, img_list):
img_num = len(img_list)
# Calculate the aspect ratio of all text bars
width_list = []
for img in img_list:
width_list.append(img.shape[1] / float(img.shape[0]))
# Sorting can speed up the recognition process
indices = np.argsort(np.array(width_list))
# rec_res = []
rec_res = [['', 0.0]] * img_num
batch_num = self.rec_batch_num
elapse = 0
for beg_img_no in range(0, img_num, batch_num):
end_img_no = min(img_num, beg_img_no + batch_num)
norm_img_batch = []
max_wh_ratio = 0
for ino in range(beg_img_no, end_img_no):
# h, w = img_list[ino].shape[0:2]
h, w = img_list[indices[ino]].shape[0:2]
wh_ratio = w * 1.0 / h
max_wh_ratio = max(max_wh_ratio, wh_ratio)
for ino in range(beg_img_no, end_img_no):
# norm_img = self.resize_norm_img(img_list[ino], max_wh_ratio)
norm_img = self.resize_norm_img(img_list[indices[ino]],
max_wh_ratio)
norm_img = norm_img[np.newaxis, :]
norm_img_batch.append(norm_img)
norm_img_batch = np.concatenate(norm_img_batch)
norm_img_batch = norm_img_batch.copy()
starttime = time.time()
input_dict = {}
input_dict[self.input_tensor.name] = norm_img_batch
outputs = self.predictor.run(self.output_tensors, input_dict)
preds = outputs[0]
rec_result = self.postprocess_op(preds)
for rno in range(len(rec_result)):
rec_res[indices[beg_img_no + rno]] = rec_result[rno]
elapse += time.time() - starttime
return rec_res, elapse
def main(args):
image_file_list = get_image_file_list(args.image_dir)
text_recognizer = TextRecognizer(args)
valid_image_file_list = []
img_list = []
for image_file in image_file_list:
img, flag = check_and_read_gif(image_file)
if not flag:
img = cv2.imread(image_file)
if img is None:
logger.info("error in loading image:{}".format(image_file))
continue
valid_image_file_list.append(image_file)
img_list.append(img)
try:
rec_res, predict_time = text_recognizer(img_list)
except:
logger.info(traceback.format_exc())
logger.info(
"ERROR!!!! \n"
"Please read the FAQ：https://github.com/PaddlePaddle/PaddleOCR#faq \n"
"If your model has tps module: "
"TPS does not support variable shape.\n"
"Please set --rec_image_shape='3,32,100' and --rec_char_type='en' ")
exit()
for ino in range(len(img_list)):
logger.info("Predicts of {}:{}".format(valid_image_file_list[ino],
rec_res[ino]))
logger.info("Total predict time for {} images, cost: {:.3f}".format(
len(img_list), predict_time))
if __name__ == "__main__":
main(utility.parse_args())
deploy/paddle2onnx/predict_system.py 已删除 100644 → 0
浏览文件 @ 1c5d1aa4
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import os
import sys
__dir__ = os.path.dirname(os.path.abspath(__file__))
sys.path.append(__dir__)
sys.path.append(os.path.abspath(os.path.join(__dir__, '../..')))
os.environ["FLAGS_allocator_strategy"] = 'auto_growth'
import cv2
import copy
import numpy as np
import time
from PIL import Image
import utility as utility
import predict_rec as predict_rec
import predict_det as predict_det
import predict_cls as predict_cls
from ppocr.utils.utility import get_image_file_list, check_and_read_gif
from ppocr.utils.logging import get_logger
from utility import draw_ocr_box_txt
logger = get_logger()
class TextSystem(object):
def __init__(self, args):
self.text_detector = predict_det.TextDetector(args)
self.text_recognizer = predict_rec.TextRecognizer(args)
self.use_angle_cls = args.use_angle_cls
self.drop_score = args.drop_score
if self.use_angle_cls:
self.text_classifier = predict_cls.TextClassifier(args)
def get_rotate_crop_image(self, img, points):
'''
img_height, img_width = img.shape[0:2]
left = int(np.min(points[:, 0]))
right = int(np.max(points[:, 0]))
top = int(np.min(points[:, 1]))
bottom = int(np.max(points[:, 1]))
img_crop = img[top:bottom, left:right, :].copy()
points[:, 0] = points[:, 0] - left
points[:, 1] = points[:, 1] - top
'''
img_crop_width = int(
max(
np.linalg.norm(points[0] - points[1]),
np.linalg.norm(points[2] - points[3])))
img_crop_height = int(
max(
np.linalg.norm(points[0] - points[3]),
np.linalg.norm(points[1] - points[2])))
pts_std = np.float32([[0, 0], [img_crop_width, 0],
[img_crop_width, img_crop_height],
[0, img_crop_height]])
M = cv2.getPerspectiveTransform(points, pts_std)
dst_img = cv2.warpPerspective(
img,
M, (img_crop_width, img_crop_height),
borderMode=cv2.BORDER_REPLICATE,
flags=cv2.INTER_CUBIC)
dst_img_height, dst_img_width = dst_img.shape[0:2]
if dst_img_height * 1.0 / dst_img_width >= 1.5:
dst_img = np.rot90(dst_img)
return dst_img
def print_draw_crop_rec_res(self, img_crop_list, rec_res):
bbox_num = len(img_crop_list)
for bno in range(bbox_num):
cv2.imwrite("./output/img_crop_%d.jpg" % bno, img_crop_list[bno])
logger.info(bno, rec_res[bno])
def __call__(self, img):
ori_im = img.copy()
dt_boxes, elapse = self.text_detector(img)
logger.info("dt_boxes num : {}, elapse : {}".format(
len(dt_boxes), elapse))
if dt_boxes is None:
return None, None
img_crop_list = []
dt_boxes = sorted_boxes(dt_boxes)
for bno in range(len(dt_boxes)):
tmp_box = copy.deepcopy(dt_boxes[bno])
img_crop = self.get_rotate_crop_image(ori_im, tmp_box)
img_crop_list.append(img_crop)
if self.use_angle_cls:
img_crop_list, angle_list, elapse = self.text_classifier(
img_crop_list)
logger.info("cls num : {}, elapse : {}".format(
len(img_crop_list), elapse))
rec_res, elapse = self.text_recognizer(img_crop_list)
logger.info("rec_res num : {}, elapse : {}".format(
len(rec_res), elapse))
# self.print_draw_crop_rec_res(img_crop_list, rec_res)
filter_boxes, filter_rec_res = [], []
for box, rec_reuslt in zip(dt_boxes, rec_res):
text, score = rec_reuslt
if score >= self.drop_score:
filter_boxes.append(box)
filter_rec_res.append(rec_reuslt)
return filter_boxes, filter_rec_res
def sorted_boxes(dt_boxes):
"""
Sort text boxes in order from top to bottom, left to right
args:
dt_boxes(array):detected text boxes with shape [4, 2]
return:
sorted boxes(array) with shape [4, 2]
"""
num_boxes = dt_boxes.shape[0]
sorted_boxes = sorted(dt_boxes, key=lambda x: (x[0][1], x[0][0]))
_boxes = list(sorted_boxes)
for i in range(num_boxes - 1):
if abs(_boxes[i + 1][0][1] - _boxes[i][0][1]) < 10 and \
(_boxes[i + 1][0][0] < _boxes[i][0][0]):
tmp = _boxes[i]
_boxes[i] = _boxes[i + 1]
_boxes[i + 1] = tmp
return _boxes
def main(args):
image_file_list = get_image_file_list(args.image_dir)
text_sys = TextSystem(args)
is_visualize = True
font_path = args.vis_font_path
drop_score = args.drop_score
for image_file in image_file_list:
img, flag = check_and_read_gif(image_file)
if not flag:
img = cv2.imread(image_file)
#img = utility.resize_img(img, 640)
#img = utility.padding_img(img, [640,640])
print(img.shape)
if img is None:
logger.info("error in loading image:{}".format(image_file))
continue
starttime = time.time()
dt_boxes, rec_res = text_sys(img)
elapse = time.time() - starttime
logger.info("Predict time of %s: %.3fs" % (image_file, elapse))
for text, score in rec_res:
logger.info("{}, {:.3f}".format(text, score))
if is_visualize:
image = Image.fromarray(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
boxes = dt_boxes
txts = [rec_res[i][0] for i in range(len(rec_res))]
scores = [rec_res[i][1] for i in range(len(rec_res))]
draw_img = draw_ocr_box_txt(
image,
boxes,
txts,
scores,
drop_score=drop_score,
font_path=font_path)
draw_img_save = "./results/"
if not os.path.exists(draw_img_save):
os.makedirs(draw_img_save)
cv2.imwrite(
os.path.join(draw_img_save, os.path.basename(image_file)),
draw_img[:, :, ::-1])
logger.info("The visualized image saved in {}".format(
os.path.join(draw_img_save, os.path.basename(image_file))))
if __name__ == "__main__":
main(utility.parse_args())
\ No newline at end of file
deploy/paddle2onnx/readme.md 0 → 100644
浏览文件 @ 3f5ffc32
# paddle2onnx 模型转化与预测
本章节介绍 PaddleOCR 模型如何转化为 ONNX 模型，并基于 ONNX 引擎预测。
## 1. 环境准备
需要准备 Paddle2ONNX 模型转化环境，和 ONNX 模型预测环境
### Paddle2ONNX
paddle2onnx 支持将 PaddlePaddle 模型格式转化到 ONNX 模型格式，算子目前稳定支持导出 ONNX Opset 9~11，部分Paddle算子支持更低的ONNX Opset转换。
更多细节可参考 [Paddle2ONNX](https://github.com/PaddlePaddle/Paddle2ONNX/blob/develop/README_zh.md)
- 安装 Paddle2ONNX
```
python3.7 -m pip install paddle2onnx
```
- 安装 ONNX
```
# 建议安装 1.4.0 版本，可根据环境更换版本号
python3.7 -m pip install onnxruntime==1.4.0
```
- Paddle 模型下载
有两种方式获取Paddle静态图模型：在 [model_list](../../doc/doc_ch/models_list.md) 中下载PaddleOCR提供的预测模型；
参考[模型导出说明](../../doc/doc_ch/inference.md#训练模型转inference模型)把训练好的权重转为 inference_model。
以 ppocr 检测模型为例：
```
wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_infer.tar
cd ./inference && tar xf ch_ppocr_mobile_v2.0_det_infer.tar && cd ..
```
## 2. 模型转换
使用 Paddle2ONNX 将Paddle静态图模型转换为ONNX模型格式：
```
paddle2onnx --model_dir=./inference/ch_ppocr_mobile_v2.0_det_infer/ \
--model_filename=inference.pdmodel \
--params_filename=inference.pdiparams \
--save_file=./inference/det_mobile_onnx/model.onnx \
--opset_version=10 \
--enable_onnx_checker=True
```
执行完毕后，ONNX模型会被保存在 `./inference/det_mobile_onnx/` 路径下
## 3. onnx 预测
以检测模型为例，使用onnx预测可执行如下命令：
```
python3.7 ../../tools/infer/predict_det.py --use_gpu=False --use_onnx=True \
--det_model_dir=./inference/det_mobile_onnx/model.onnx \
--image_dir=../../doc/imgs/1.jpg
```
执行命令后在终端会打印出预测的检测框坐标，并在 `./inference_results/` 下保存可视化结果。
```
root INFO: 1.jpg [[[291, 295], [334, 292], [348, 844], [305, 847]], [[344, 296], [379, 294], [387, 669], [353, 671]]]
The predict time of ../../doc/imgs/1.jpg: 0.06162881851196289
The visualized image saved in ./inference_results/det_res_1.jpg
```
* 注意：ONNX暂时不支持变长预测，因为需要将输入resize到固定输入，预测结果可能与直接使用Paddle预测有细微不同。
deploy/paddle2onnx/utility.py 已删除 100644 → 0
浏览文件 @ 1c5d1aa4
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import argparse
import os
import sys
import cv2
import numpy as np
import json
from PIL import Image, ImageDraw, ImageFont
import math
import onnxruntime as ort
def parse_args():
def str2bool(v):
return v.lower() in ("true", "t", "1")
parser = argparse.ArgumentParser()
# params for prediction engine
parser.add_argument("--use_gpu", type=str2bool, default=True)
parser.add_argument("--ir_optim", type=str2bool, default=True)
parser.add_argument("--use_tensorrt", type=str2bool, default=False)
parser.add_argument("--use_fp16", type=str2bool, default=False)
parser.add_argument("--gpu_mem", type=int, default=500)
# params for text detector
parser.add_argument("--image_dir", type=str)
parser.add_argument("--det_algorithm", type=str, default='DB')
parser.add_argument("--det_model_dir", type=str)
parser.add_argument("--det_limit_side_len", type=float, default=960)
parser.add_argument("--det_limit_type", type=str, default='max')
# DB parmas
parser.add_argument("--det_db_thresh", type=float, default=0.3)
parser.add_argument("--det_db_box_thresh", type=float, default=0.5)
parser.add_argument("--det_db_unclip_ratio", type=float, default=1.6)
parser.add_argument("--max_batch_size", type=int, default=10)
# EAST parmas
parser.add_argument("--det_east_score_thresh", type=float, default=0.8)
parser.add_argument("--det_east_cover_thresh", type=float, default=0.1)
parser.add_argument("--det_east_nms_thresh", type=float, default=0.2)
# SAST parmas
parser.add_argument("--det_sast_score_thresh", type=float, default=0.5)
parser.add_argument("--det_sast_nms_thresh", type=float, default=0.2)
parser.add_argument("--det_sast_polygon", type=bool, default=False)
# params for text recognizer
parser.add_argument("--rec_algorithm", type=str, default='CRNN')
parser.add_argument("--rec_model_dir", type=str)
parser.add_argument("--rec_image_shape", type=str, default="3, 32, 100")
parser.add_argument("--rec_char_type", type=str, default='ch')
parser.add_argument("--rec_batch_num", type=int, default=6)
parser.add_argument("--max_text_length", type=int, default=25)
parser.add_argument(
"--rec_char_dict_path",
type=str,
default="./ppocr/utils/ppocr_keys_v1.txt")
parser.add_argument("--use_space_char", type=str2bool, default=True)
parser.add_argument(
"--vis_font_path", type=str, default="./doc/simfang.ttf")
parser.add_argument("--drop_score", type=float, default=0.5)
# params for text classifier
parser.add_argument("--use_angle_cls", type=str2bool, default=False)
parser.add_argument("--cls_model_dir", type=str)
parser.add_argument("--cls_image_shape", type=str, default="3, 48, 192")
parser.add_argument("--label_list", type=list, default=['0', '180'])
parser.add_argument("--cls_batch_num", type=int, default=6)
parser.add_argument("--cls_thresh", type=float, default=0.9)
parser.add_argument("--enable_mkldnn", type=str2bool, default=False)
parser.add_argument("--use_pdserving", type=str2bool, default=False)
return parser.parse_args()
def create_predictor(args, mode, logger):
if mode == "det":
model_dir = args.det_model_dir
elif mode == 'cls':
model_dir = args.cls_model_dir
else:
model_dir = args.rec_model_dir
if model_dir is None:
logger.info("not find {} model file path {}".format(mode, model_dir))
sys.exit(0)
model_file_path = model_dir
if not os.path.exists(model_file_path):
logger.info("not find model file path {}".format(model_file_path))
sys.exit(0)
sess = ort.InferenceSession(model_file_path)
return sess, sess.get_inputs()[0], None
def draw_text_det_res(dt_boxes, img_path):
src_im = cv2.imread(img_path)
for box in dt_boxes:
box = np.array(box).astype(np.int32).reshape(-1, 2)
cv2.polylines(src_im, [box], True, color=(255, 255, 0), thickness=2)
return src_im
def resize_img(img, input_size=600):
"""
resize img and limit the longest side of the image to input_size
"""
img = np.array(img)
im_shape = img.shape
im_size_max = np.max(im_shape[0:2])
im_scale = float(input_size) / float(im_size_max)
img = cv2.resize(img, None, None, fx=im_scale, fy=im_scale)
return img
def padding_img(img, input_size=[640, 640]):
padding_im_h = input_size[0]
padding_im_w = input_size[1]
im_c = img.shape[2]
im_h = img.shape[0]
im_w = img.shape[1]
padding_im = np.zeros(
(padding_im_h, padding_im_w, im_c), dtype=np.float32) + 255
padding_im[:im_h, :im_w, :] = img
return padding_im
def draw_ocr(image,
boxes,
txts=None,
scores=None,
drop_score=0.5,
font_path="./doc/simfang.ttf"):
"""
Visualize the results of OCR detection and recognition
args:
image(Image|array): RGB image
boxes(list): boxes with shape(N, 4, 2)
txts(list): the texts
scores(list): txxs corresponding scores
drop_score(float): only scores greater than drop_threshold will be visualized
font_path: the path of font which is used to draw text
return(array):
the visualized img
"""
if scores is None:
scores = [1] * len(boxes)
box_num = len(boxes)
for i in range(box_num):
if scores is not None and (scores[i] < drop_score or
math.isnan(scores[i])):
continue
box = np.reshape(np.array(boxes[i]), [-1, 1, 2]).astype(np.int64)
image = cv2.polylines(np.array(image), [box], True, (255, 0, 0), 2)
if txts is not None:
img = np.array(resize_img(image, input_size=600))
txt_img = text_visual(
txts,
scores,
img_h=img.shape[0],
img_w=600,
threshold=drop_score,
font_path=font_path)
img = np.concatenate([np.array(img), np.array(txt_img)], axis=1)
return img
return image
def draw_ocr_box_txt(image,
boxes,
txts,
scores=None,
drop_score=0.5,
font_path="./doc/fonts/simfang.ttf"):
h, w = image.height, image.width
img_left = image.copy()
img_right = Image.new('RGB', (w, h), (255, 255, 255))
import random
random.seed(0)
draw_left = ImageDraw.Draw(img_left)
draw_right = ImageDraw.Draw(img_right)
for idx, (box, txt) in enumerate(zip(boxes, txts)):
if scores is not None and scores[idx] < drop_score:
continue
color = (random.randint(0, 255), random.randint(0, 255),
random.randint(0, 255))
draw_left.polygon(box, fill=color)
draw_right.polygon(
[
box[0][0], box[0][1], box[1][0], box[1][1], box[2][0],
box[2][1], box[3][0], box[3][1]
],
outline=color)
box_height = math.sqrt((box[0][0] - box[3][0])**2 + (box[0][1] - box[3][
1])**2)
box_width = math.sqrt((box[0][0] - box[1][0])**2 + (box[0][1] - box[1][
1])**2)
if box_height > 2 * box_width:
font_size = max(int(box_width * 0.9), 10)
font = ImageFont.truetype(font_path, font_size, encoding="utf-8")
cur_y = box[0][1]
for c in txt:
char_size = font.getsize(c)
draw_right.text(
(box[0][0] + 3, cur_y), c, fill=(0, 0, 0), font=font)
cur_y += char_size[1]
else:
font_size = max(int(box_height * 0.8), 10)
font = ImageFont.truetype(font_path, font_size, encoding="utf-8")
draw_right.text(
[box[0][0], box[0][1]], txt, fill=(0, 0, 0), font=font)
img_left = Image.blend(image, img_left, 0.5)
img_show = Image.new('RGB', (w * 2, h), (255, 255, 255))
img_show.paste(img_left, (0, 0, w, h))
img_show.paste(img_right, (w, 0, w * 2, h))
return np.array(img_show)
def str_count(s):
"""
Count the number of Chinese characters,
a single English character and a single number
equal to half the length of Chinese characters.
args:
s(string): the input of string
return(int):
the number of Chinese characters
"""
import string
count_zh = count_pu = 0
s_len = len(s)
en_dg_count = 0
for c in s:
if c in string.ascii_letters or c.isdigit() or c.isspace():
en_dg_count += 1
elif c.isalpha():
count_zh += 1
else:
count_pu += 1
return s_len - math.ceil(en_dg_count / 2)
def text_visual(texts,
scores,
img_h=400,
img_w=600,
threshold=0.,
font_path="./doc/simfang.ttf"):
"""
create new blank img and draw txt on it
args:
texts(list): the text will be draw
scores(list|None): corresponding score of each txt
img_h(int): the height of blank img
img_w(int): the width of blank img
font_path: the path of font which is used to draw text
return(array):
"""
if scores is not None:
assert len(texts) == len(
scores), "The number of txts and corresponding scores must match"
def create_blank_img():
blank_img = np.ones(shape=[img_h, img_w], dtype=np.int8) * 255
blank_img[:, img_w - 1:] = 0
blank_img = Image.fromarray(blank_img).convert("RGB")
draw_txt = ImageDraw.Draw(blank_img)
return blank_img, draw_txt
blank_img, draw_txt = create_blank_img()
font_size = 20
txt_color = (0, 0, 0)
font = ImageFont.truetype(font_path, font_size, encoding="utf-8")
gap = font_size + 5
txt_img_list = []
count, index = 1, 0
for idx, txt in enumerate(texts):
index += 1
if scores[idx] < threshold or math.isnan(scores[idx]):
index -= 1
continue
first_line = True
while str_count(txt) >= img_w // font_size - 4:
tmp = txt
txt = tmp[:img_w // font_size - 4]
if first_line:
new_txt = str(index) + ': ' + txt
first_line = False
else:
new_txt = ' ' + txt
draw_txt.text((0, gap * count), new_txt, txt_color, font=font)
txt = tmp[img_w // font_size - 4:]
if count >= img_h // gap - 1:
txt_img_list.append(np.array(blank_img))
blank_img, draw_txt = create_blank_img()
count = 0
count += 1
if first_line:
new_txt = str(index) + ': ' + txt + ' ' + '%.3f' % (scores[idx])
else:
new_txt = " " + txt + " " + '%.3f' % (scores[idx])
draw_txt.text((0, gap * count), new_txt, txt_color, font=font)
# whether add new blank img or not
if count >= img_h // gap - 1 and idx + 1 < len(texts):
txt_img_list.append(np.array(blank_img))
blank_img, draw_txt = create_blank_img()
count = 0
count += 1
txt_img_list.append(np.array(blank_img))
if len(txt_img_list) == 1:
blank_img = np.array(txt_img_list[0])
else:
blank_img = np.concatenate(txt_img_list, axis=1)
return np.array(blank_img)
def base64_to_cv2(b64str):
import base64
data = base64.b64decode(b64str.encode('utf8'))
data = np.fromstring(data, np.uint8)
data = cv2.imdecode(data, cv2.IMREAD_COLOR)
return data
def draw_boxes(image, boxes, scores=None, drop_score=0.5):
if scores is None:
scores = [1] * len(boxes)
for (box, score) in zip(boxes, scores):
if score < drop_score:
continue
box = np.reshape(np.array(box), [-1, 1, 2]).astype(np.int64)
image = cv2.polylines(np.array(image), [box], True, (255, 0, 0), 2)
return image
if __name__ == '__main__':
test_img = "./doc/test_v2"
predict_txt = "./doc/predict.txt"
f = open(predict_txt, 'r')
data = f.readlines()
img_path, anno = data[0].strip().split('\t')
img_name = os.path.basename(img_path)
img_path = os.path.join(test_img, img_name)
image = Image.open(img_path)
data = json.loads(anno)
boxes, txts, scores = [], [], []
for dic in data:
boxes.append(dic['points'])
txts.append(dic['transcription'])
scores.append(round(dic['scores'], 3))
new_img = draw_ocr(image, boxes, txts, scores)
cv2.imwrite(img_name, new_img)
test_tipc/configs/ppocr_det_mobile_params.txt
浏览文件 @ 3f5ffc32
......@@ -116,7 +116,7 @@ infer_model:./models/ch_ppocr_mobile_v2.0_det_opt.nb|./models/ch_ppocr_mobile_v2
--save_file:./inference/det_mobile_onnx/model.onnx
--opset_version:10
--enable_onnx_checker:True
inference:deploy/paddle2onnx/predict_det.py
inference:tools/infer/predict_det.py
--use_gpu:False
--det_model_dir:
--image_dir:./inference/ch_det_data_50/all-sum-510/
\ No newline at end of file
test_tipc/test_paddle2onnx.sh
浏览文件 @ 3f5ffc32
......@@ -11,7 +11,7 @@ python=$(func_parser_value "${lines[2]}")
# parser params
dataline=$(awk 'NR==111, NR==122{print}' $FILENAME)
dataline=$(awk 'NR==111, NR==123{print}' $FILENAME)
IFS=$'\n'
lines=(${dataline})
......@@ -63,7 +63,7 @@ function func_paddle2onnx(){
set_gpu=$(func_set_params "${use_gpu_key}" "${use_gpu_value}")
set_model_dir=$(func_set_params "${det_model_key}" "${save_file_value}")
set_img_dir=$(func_set_params "${image_dir_key}" "${image_dir_value}")
infer_model_cmd="${python} ${inference_py} ${set_gpu} ${set_img_dir} ${set_model_dir} > ${_save_log_path} 2>&1 "
infer_model_cmd="${python} ${inference_py} ${set_gpu} ${set_img_dir} ${set_model_dir} --use_onnx=True > ${_save_log_path} 2>&1 "
eval $infer_model_cmd
status_check $last_status "${infer_model_cmd}" "${status_log}"
}
......
tools/infer/predict_cls.py
浏览文件 @ 3f5ffc32
......@@ -47,6 +47,7 @@ class TextClassifier(object):
self.postprocess_op = build_post_process(postprocess_params)
self.predictor, self.input_tensor, self.output_tensors, _ = \
utility.create_predictor(args, 'cls', logger)
self.use_onnx = args.use_onnx
def resize_norm_img(self, img):
imgC, imgH, imgW = self.cls_image_shape
......@@ -100,10 +101,16 @@ class TextClassifier(object):
norm_img_batch = np.concatenate(norm_img_batch)
norm_img_batch = norm_img_batch.copy()
self.input_tensor.copy_from_cpu(norm_img_batch)
self.predictor.run()
prob_out = self.output_tensors[0].copy_to_cpu()
self.predictor.try_shrink_memory()
if self.use_onnx:
input_dict = {}
input_dict[self.input_tensor.name] = norm_img_batch
outputs = self.predictor.run(self.output_tensors, input_dict)
prob_out = outputs[0]
else:
self.input_tensor.copy_from_cpu(norm_img_batch)
self.predictor.run()
prob_out = self.output_tensors[0].copy_to_cpu()
self.predictor.try_shrink_memory()
cls_result = self.postprocess_op(prob_out)
elapse += time.time() - starttime
for rno in range(len(cls_result)):
......
tools/infer/predict_det.py
浏览文件 @ 3f5ffc32
......@@ -38,6 +38,7 @@ class TextDetector(object):
def __init__(self, args):
self.args = args
self.det_algorithm = args.det_algorithm
self.use_onnx = args.use_onnx
pre_process_list = [{
'DetResizeForTest': {
'limit_side_len': args.det_limit_side_len,
......@@ -100,7 +101,12 @@ class TextDetector(object):
else:
logger.info("unknown det_algorithm:{}".format(self.det_algorithm))
sys.exit(0)
if self.use_onnx:
pre_process_list[0] = {
'DetResizeForTest': {
'image_shape': [640, 640]
}
}
self.preprocess_op = create_operators(pre_process_list)
self.postprocess_op = build_post_process(postprocess_params)
self.predictor, self.input_tensor, self.output_tensors, self.config = utility.create_predictor(
......@@ -198,15 +204,19 @@ class TextDetector(object):
if self.args.benchmark:
self.autolog.times.stamp()
self.input_tensor.copy_from_cpu(img)
self.predictor.run()
outputs = []
for output_tensor in self.output_tensors:
output = output_tensor.copy_to_cpu()
outputs.append(output)
if self.args.benchmark:
self.autolog.times.stamp()
if self.use_onnx:
input_dict = {}
input_dict[self.input_tensor.name] = img
outputs = self.predictor.run(self.output_tensors, input_dict)
else:
self.input_tensor.copy_from_cpu(img)
self.predictor.run()
outputs = []
for output_tensor in self.output_tensors:
output = output_tensor.copy_to_cpu()
outputs.append(output)
if self.args.benchmark:
self.autolog.times.stamp()
preds = {}
if self.det_algorithm == "EAST":
......
tools/infer/predict_rec.py
浏览文件 @ 3f5ffc32
......@@ -73,6 +73,7 @@ class TextRecognizer(object):
self.predictor, self.input_tensor, self.output_tensors, self.config = \
utility.create_predictor(args, 'rec', logger)
self.benchmark = args.benchmark
self.use_onnx = args.use_onnx
if args.benchmark:
import auto_log
pid = os.getpid()
......@@ -108,6 +109,8 @@ class TextRecognizer(object):
assert imgC == img.shape[2]
max_wh_ratio = max(max_wh_ratio, imgW / imgH)
imgW = int((32 * max_wh_ratio))
if self.use_onnx:
imgW = 100
h, w = img.shape[:2]
ratio = w / float(h)
if math.ceil(imgH * ratio) > imgW:
......@@ -297,51 +300,72 @@ class TextRecognizer(object):
gsrm_slf_attn_bias1_list,
gsrm_slf_attn_bias2_list,
]
input_names = self.predictor.get_input_names()
for i in range(len(input_names)):
input_tensor = self.predictor.get_input_handle(input_names[
i])
input_tensor.copy_from_cpu(inputs[i])
self.predictor.run()
outputs = []
for output_tensor in self.output_tensors:
output = output_tensor.copy_to_cpu()
outputs.append(output)
if self.benchmark:
self.autolog.times.stamp()
preds = {"predict": outputs[2]}
if self.use_onnx:
input_dict = {}
input_dict[self.input_tensor.name] = norm_img_batch
outputs = self.predictor.run(self.output_tensors,
input_dict)
preds = {"predict": outputs[2]}
else:
input_names = self.predictor.get_input_names()
for i in range(len(input_names)):
input_tensor = self.predictor.get_input_handle(
input_names[i])
input_tensor.copy_from_cpu(inputs[i])
self.predictor.run()
outputs = []
for output_tensor in self.output_tensors:
output = output_tensor.copy_to_cpu()
outputs.append(output)
if self.benchmark:
self.autolog.times.stamp()
preds = {"predict": outputs[2]}
elif self.rec_algorithm == "SAR":
valid_ratios = np.concatenate(valid_ratios)
inputs = [
norm_img_batch,
valid_ratios,
]
input_names = self.predictor.get_input_names()
for i in range(len(input_names)):
input_tensor = self.predictor.get_input_handle(input_names[
i])
input_tensor.copy_from_cpu(inputs[i])
self.predictor.run()
outputs = []
for output_tensor in self.output_tensors:
output = output_tensor.copy_to_cpu()
outputs.append(output)
if self.benchmark:
self.autolog.times.stamp()
preds = outputs[0]
else:
self.input_tensor.copy_from_cpu(norm_img_batch)
self.predictor.run()
outputs = []
for output_tensor in self.output_tensors:
output = output_tensor.copy_to_cpu()
outputs.append(output)
if self.benchmark:
self.autolog.times.stamp()
if len(outputs) != 1:
preds = outputs
if self.use_onnx:
input_dict = {}
input_dict[self.input_tensor.name] = norm_img_batch
outputs = self.predictor.run(self.output_tensors,
input_dict)
preds = outputs[0]
else:
input_names = self.predictor.get_input_names()
for i in range(len(input_names)):
input_tensor = self.predictor.get_input_handle(
input_names[i])
input_tensor.copy_from_cpu(inputs[i])
self.predictor.run()
outputs = []
for output_tensor in self.output_tensors:
output = output_tensor.copy_to_cpu()
outputs.append(output)
if self.benchmark:
self.autolog.times.stamp()
preds = outputs[0]
else:
if self.use_onnx:
input_dict = {}
input_dict[self.input_tensor.name] = norm_img_batch
outputs = self.predictor.run(self.output_tensors,
input_dict)
preds = outputs[0]
else:
self.input_tensor.copy_from_cpu(norm_img_batch)
self.predictor.run()
outputs = []
for output_tensor in self.output_tensors:
output = output_tensor.copy_to_cpu()
outputs.append(output)
if self.benchmark:
self.autolog.times.stamp()
if len(outputs) != 1:
preds = outputs
else:
preds = outputs[0]
rec_result = self.postprocess_op(preds)
for rno in range(len(rec_result)):
rec_res[indices[beg_img_no + rno]] = rec_result[rno]
......
tools/infer/utility.py
浏览文件 @ 3f5ffc32
......@@ -121,6 +121,7 @@ def init_args():
parser.add_argument("--save_log_path", type=str, default="./log_output/")
parser.add_argument("--show_log", type=str2bool, default=True)
parser.add_argument("--use_onnx", type=str2bool, default=False)
return parser
......@@ -144,152 +145,163 @@ def create_predictor(args, mode, logger):
if model_dir is None:
logger.info("not find {} model file path {}".format(mode, model_dir))
sys.exit(0)
model_file_path = model_dir + "/inference.pdmodel"
params_file_path = model_dir + "/inference.pdiparams"
if not os.path.exists(model_file_path):
raise ValueError("not find model file path {}".format(model_file_path))
if not os.path.exists(params_file_path):
raise ValueError("not find params file path {}".format(
params_file_path))
config = inference.Config(model_file_path, params_file_path)
if hasattr(args, 'precision'):
if args.precision == "fp16" and args.use_tensorrt:
precision = inference.PrecisionType.Half
elif args.precision == "int8":
precision = inference.PrecisionType.Int8
else:
precision = inference.PrecisionType.Float32
if args.use_onnx:
import onnxruntime as ort
model_file_path = model_dir
if not os.path.exists(model_file_path):
raise ValueError("not find model file path {}".format(
model_file_path))
sess = ort.InferenceSession(model_file_path)
return sess, sess.get_inputs()[0], None, None
else:
precision = inference.PrecisionType.Float32
if args.use_gpu:
gpu_id = get_infer_gpuid()
if gpu_id is None:
raise ValueError(
"Not found GPU in current device. Please check your device or set args.use_gpu as False"
)
config.enable_use_gpu(args.gpu_mem, 0)
if args.use_tensorrt:
config.enable_tensorrt_engine(
precision_mode=precision,
max_batch_size=args.max_batch_size,
min_subgraph_size=args.min_subgraph_size)
# skip the minmum trt subgraph
if mode == "det":
min_input_shape = {
"x": [1, 3, 50, 50],
"conv2d_92.tmp_0": [1, 120, 20, 20],
"conv2d_91.tmp_0": [1, 24, 10, 10],
"conv2d_59.tmp_0": [1, 96, 20, 20],
"nearest_interp_v2_1.tmp_0": [1, 256, 10, 10],
"nearest_interp_v2_2.tmp_0": [1, 256, 20, 20],
"conv2d_124.tmp_0": [1, 256, 20, 20],
"nearest_interp_v2_3.tmp_0": [1, 64, 20, 20],
"nearest_interp_v2_4.tmp_0": [1, 64, 20, 20],
"nearest_interp_v2_5.tmp_0": [1, 64, 20, 20],
"elementwise_add_7": [1, 56, 2, 2],
"nearest_interp_v2_0.tmp_0": [1, 256, 2, 2]
}
max_input_shape = {
"x": [1, 3, 2000, 2000],
"conv2d_92.tmp_0": [1, 120, 400, 400],
"conv2d_91.tmp_0": [1, 24, 200, 200],
"conv2d_59.tmp_0": [1, 96, 400, 400],
"nearest_interp_v2_1.tmp_0": [1, 256, 200, 200],
"conv2d_124.tmp_0": [1, 256, 400, 400],
"nearest_interp_v2_2.tmp_0": [1, 256, 400, 400],
"nearest_interp_v2_3.tmp_0": [1, 64, 400, 400],
"nearest_interp_v2_4.tmp_0": [1, 64, 400, 400],
"nearest_interp_v2_5.tmp_0": [1, 64, 400, 400],
"elementwise_add_7": [1, 56, 400, 400],
"nearest_interp_v2_0.tmp_0": [1, 256, 400, 400]
}
opt_input_shape = {
"x": [1, 3, 640, 640],
"conv2d_92.tmp_0": [1, 120, 160, 160],
"conv2d_91.tmp_0": [1, 24, 80, 80],
"conv2d_59.tmp_0": [1, 96, 160, 160],
"nearest_interp_v2_1.tmp_0": [1, 256, 80, 80],
"nearest_interp_v2_2.tmp_0": [1, 256, 160, 160],
"conv2d_124.tmp_0": [1, 256, 160, 160],
"nearest_interp_v2_3.tmp_0": [1, 64, 160, 160],
"nearest_interp_v2_4.tmp_0": [1, 64, 160, 160],
"nearest_interp_v2_5.tmp_0": [1, 64, 160, 160],
"elementwise_add_7": [1, 56, 40, 40],
"nearest_interp_v2_0.tmp_0": [1, 256, 40, 40]
}
min_pact_shape = {
"nearest_interp_v2_26.tmp_0": [1, 256, 20, 20],
"nearest_interp_v2_27.tmp_0": [1, 64, 20, 20],
"nearest_interp_v2_28.tmp_0": [1, 64, 20, 20],
"nearest_interp_v2_29.tmp_0": [1, 64, 20, 20]
}
max_pact_shape = {
"nearest_interp_v2_26.tmp_0": [1, 256, 400, 400],
"nearest_interp_v2_27.tmp_0": [1, 64, 400, 400],
"nearest_interp_v2_28.tmp_0": [1, 64, 400, 400],
"nearest_interp_v2_29.tmp_0": [1, 64, 400, 400]
}
opt_pact_shape = {
"nearest_interp_v2_26.tmp_0": [1, 256, 160, 160],
"nearest_interp_v2_27.tmp_0": [1, 64, 160, 160],
"nearest_interp_v2_28.tmp_0": [1, 64, 160, 160],
"nearest_interp_v2_29.tmp_0": [1, 64, 160, 160]
}
min_input_shape.update(min_pact_shape)
max_input_shape.update(max_pact_shape)
opt_input_shape.update(opt_pact_shape)
elif mode == "rec":
min_input_shape = {"x": [1, 3, 32, 10]}
max_input_shape = {"x": [args.rec_batch_num, 3, 32, 2000]}
opt_input_shape = {"x": [args.rec_batch_num, 3, 32, 320]}
elif mode == "cls":
min_input_shape = {"x": [1, 3, 48, 10]}
max_input_shape = {"x": [args.rec_batch_num, 3, 48, 2000]}
opt_input_shape = {"x": [args.rec_batch_num, 3, 48, 320]}
model_file_path = model_dir + "/inference.pdmodel"
params_file_path = model_dir + "/inference.pdiparams"
if not os.path.exists(model_file_path):
raise ValueError("not find model file path {}".format(
model_file_path))
if not os.path.exists(params_file_path):
raise ValueError("not find params file path {}".format(
params_file_path))
config = inference.Config(model_file_path, params_file_path)
if hasattr(args, 'precision'):
if args.precision == "fp16" and args.use_tensorrt:
precision = inference.PrecisionType.Half
elif args.precision == "int8":
precision = inference.PrecisionType.Int8
else:
precision = inference.PrecisionType.Float32
else:
min_input_shape = {"x": [1, 3, 10, 10]}
max_input_shape = {"x": [1, 3, 1000, 1000]}
opt_input_shape = {"x": [1, 3, 500, 500]}
config.set_trt_dynamic_shape_info(min_input_shape, max_input_shape,
opt_input_shape)
precision = inference.PrecisionType.Float32
if args.use_gpu:
gpu_id = get_infer_gpuid()
if gpu_id is None:
raise ValueError(
"Not found GPU in current device. Please check your device or set args.use_gpu as False"
)
config.enable_use_gpu(args.gpu_mem, 0)
if args.use_tensorrt:
config.enable_tensorrt_engine(
precision_mode=precision,
max_batch_size=args.max_batch_size,
min_subgraph_size=args.min_subgraph_size)
# skip the minmum trt subgraph
if mode == "det":
min_input_shape = {
"x": [1, 3, 50, 50],
"conv2d_92.tmp_0": [1, 120, 20, 20],
"conv2d_91.tmp_0": [1, 24, 10, 10],
"conv2d_59.tmp_0": [1, 96, 20, 20],
"nearest_interp_v2_1.tmp_0": [1, 256, 10, 10],
"nearest_interp_v2_2.tmp_0": [1, 256, 20, 20],
"conv2d_124.tmp_0": [1, 256, 20, 20],
"nearest_interp_v2_3.tmp_0": [1, 64, 20, 20],
"nearest_interp_v2_4.tmp_0": [1, 64, 20, 20],
"nearest_interp_v2_5.tmp_0": [1, 64, 20, 20],
"elementwise_add_7": [1, 56, 2, 2],
"nearest_interp_v2_0.tmp_0": [1, 256, 2, 2]
}
max_input_shape = {
"x": [1, 3, 2000, 2000],
"conv2d_92.tmp_0": [1, 120, 400, 400],
"conv2d_91.tmp_0": [1, 24, 200, 200],
"conv2d_59.tmp_0": [1, 96, 400, 400],
"nearest_interp_v2_1.tmp_0": [1, 256, 200, 200],
"conv2d_124.tmp_0": [1, 256, 400, 400],
"nearest_interp_v2_2.tmp_0": [1, 256, 400, 400],
"nearest_interp_v2_3.tmp_0": [1, 64, 400, 400],
"nearest_interp_v2_4.tmp_0": [1, 64, 400, 400],
"nearest_interp_v2_5.tmp_0": [1, 64, 400, 400],
"elementwise_add_7": [1, 56, 400, 400],
"nearest_interp_v2_0.tmp_0": [1, 256, 400, 400]
}
opt_input_shape = {
"x": [1, 3, 640, 640],
"conv2d_92.tmp_0": [1, 120, 160, 160],
"conv2d_91.tmp_0": [1, 24, 80, 80],
"conv2d_59.tmp_0": [1, 96, 160, 160],
"nearest_interp_v2_1.tmp_0": [1, 256, 80, 80],
"nearest_interp_v2_2.tmp_0": [1, 256, 160, 160],
"conv2d_124.tmp_0": [1, 256, 160, 160],
"nearest_interp_v2_3.tmp_0": [1, 64, 160, 160],
"nearest_interp_v2_4.tmp_0": [1, 64, 160, 160],
"nearest_interp_v2_5.tmp_0": [1, 64, 160, 160],
"elementwise_add_7": [1, 56, 40, 40],
"nearest_interp_v2_0.tmp_0": [1, 256, 40, 40]
}
min_pact_shape = {
"nearest_interp_v2_26.tmp_0": [1, 256, 20, 20],
"nearest_interp_v2_27.tmp_0": [1, 64, 20, 20],
"nearest_interp_v2_28.tmp_0": [1, 64, 20, 20],
"nearest_interp_v2_29.tmp_0": [1, 64, 20, 20]
}
max_pact_shape = {
"nearest_interp_v2_26.tmp_0": [1, 256, 400, 400],
"nearest_interp_v2_27.tmp_0": [1, 64, 400, 400],
"nearest_interp_v2_28.tmp_0": [1, 64, 400, 400],
"nearest_interp_v2_29.tmp_0": [1, 64, 400, 400]
}
opt_pact_shape = {
"nearest_interp_v2_26.tmp_0": [1, 256, 160, 160],
"nearest_interp_v2_27.tmp_0": [1, 64, 160, 160],
"nearest_interp_v2_28.tmp_0": [1, 64, 160, 160],
"nearest_interp_v2_29.tmp_0": [1, 64, 160, 160]
}
min_input_shape.update(min_pact_shape)
max_input_shape.update(max_pact_shape)
opt_input_shape.update(opt_pact_shape)
elif mode == "rec":
min_input_shape = {"x": [1, 3, 32, 10]}
max_input_shape = {"x": [args.rec_batch_num, 3, 32, 2000]}
opt_input_shape = {"x": [args.rec_batch_num, 3, 32, 320]}
elif mode == "cls":
min_input_shape = {"x": [1, 3, 48, 10]}
max_input_shape = {"x": [args.rec_batch_num, 3, 48, 2000]}
opt_input_shape = {"x": [args.rec_batch_num, 3, 48, 320]}
else:
min_input_shape = {"x": [1, 3, 10, 10]}
max_input_shape = {"x": [1, 3, 1000, 1000]}
opt_input_shape = {"x": [1, 3, 500, 500]}
config.set_trt_dynamic_shape_info(min_input_shape, max_input_shape,
opt_input_shape)
else:
config.disable_gpu()
if hasattr(args, "cpu_threads"):
config.set_cpu_math_library_num_threads(args.cpu_threads)
else:
# default cpu threads as 10
config.set_cpu_math_library_num_threads(10)
if args.enable_mkldnn:
# cache 10 different shapes for mkldnn to avoid memory leak
config.set_mkldnn_cache_capacity(10)
config.enable_mkldnn()
if args.precision == "fp16":
config.enable_mkldnn_bfloat16()
# enable memory optim
config.enable_memory_optim()
config.disable_glog_info()
config.delete_pass("conv_transpose_eltwiseadd_bn_fuse_pass")
if mode == 'table':
config.delete_pass("fc_fuse_pass") # not supported for table
config.switch_use_feed_fetch_ops(False)
config.switch_ir_optim(True)
# create predictor
predictor = inference.create_predictor(config)
input_names = predictor.get_input_names()
for name in input_names:
input_tensor = predictor.get_input_handle(name)
output_names = predictor.get_output_names()
output_tensors = []
for output_name in output_names:
output_tensor = predictor.get_output_handle(output_name)
output_tensors.append(output_tensor)
return predictor, input_tensor, output_tensors, config
config.disable_gpu()
if hasattr(args, "cpu_threads"):
config.set_cpu_math_library_num_threads(args.cpu_threads)
else:
# default cpu threads as 10
config.set_cpu_math_library_num_threads(10)
if args.enable_mkldnn:
# cache 10 different shapes for mkldnn to avoid memory leak
config.set_mkldnn_cache_capacity(10)
config.enable_mkldnn()
if args.precision == "fp16":
config.enable_mkldnn_bfloat16()
# enable memory optim
config.enable_memory_optim()
config.disable_glog_info()
config.delete_pass("conv_transpose_eltwiseadd_bn_fuse_pass")
if mode == 'table':
config.delete_pass("fc_fuse_pass") # not supported for table
config.switch_use_feed_fetch_ops(False)
config.switch_ir_optim(True)
# create predictor
predictor = inference.create_predictor(config)
input_names = predictor.get_input_names()
for name in input_names:
input_tensor = predictor.get_input_handle(name)
output_names = predictor.get_output_names()
output_tensors = []
for output_name in output_names:
output_tensor = predictor.get_output_handle(output_name)
output_tensors.append(output_tensor)
return predictor, input_tensor, output_tensors, config
def get_infer_gpuid():
......
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或