From f01dbb5648a97af6a3160e5653affe5a0bb3dc4e Mon Sep 17 00:00:00 2001
From: WenmuZhou <572459439@qq.com>
Date: Mon, 20 Dec 2021 14:23:36 +0000
Subject: [PATCH 01/35] add LayoutLM ser
---
ppstructure/vqa/README.md | 16 +++++++----
ppstructure/vqa/eval_ser.py | 47 +++++++++++++++++++++---------
ppstructure/vqa/infer_re.py | 9 ++++--
ppstructure/vqa/infer_ser.py | 47 +++++++++++++++++++-----------
ppstructure/vqa/infer_ser_e2e.py | 49 +++++++++++++++++++++-----------
ppstructure/vqa/losses.py | 35 +++++++++++++++++++++++
ppstructure/vqa/train_ser.py | 48 ++++++++++++++++++++++---------
ppstructure/vqa/utils.py | 2 ++
8 files changed, 185 insertions(+), 68 deletions(-)
create mode 100644 ppstructure/vqa/losses.py
diff --git a/ppstructure/vqa/README.md b/ppstructure/vqa/README.md
index 2216950e5..51a616c43 100644
--- a/ppstructure/vqa/README.md
+++ b/ppstructure/vqa/README.md
@@ -18,12 +18,13 @@ PP-Structure 里的 DOC-VQA算法基于PaddleNLP自然语言处理算法库进
## 1 性能
-我们在 [XFUN](https://github.com/doc-analysis/XFUND) 评估数据集上对算法进行了评估，性能如下
+我们在 [XFUN](https://github.com/doc-analysis/XFUND) 的中文数据集上对算法进行了评估，性能如下
-|任务| f1 | 模型下载地址|
-|:---:|:---:| :---:|
-|SER|0.9056| [链接](https://paddleocr.bj.bcebos.com/pplayout/PP-Layout_v1.0_ser_pretrained.tar)|
-|RE|0.7113| [链接](https://paddleocr.bj.bcebos.com/pplayout/PP-Layout_v1.0_re_pretrained.tar)|
+| 模型 | 任务 | f1 | 模型下载地址 |
+|:---:|:---:|:---:| :---:|
+| LayoutXLM | RE | 0.7113 | [链接](https://paddleocr.bj.bcebos.com/pplayout/PP-Layout_v1.0_re_pretrained.tar) |
+| LayoutXLM | SER | 0.9056 | [链接](https://paddleocr.bj.bcebos.com/pplayout/PP-Layout_v1.0_ser_pretrained.tar) |
+| LayoutLM | SER | 0.78 | [链接](https://paddleocr.bj.bcebos.com/pplayout/LayoutLM_ser_pretrained.tar) |
@@ -135,6 +136,7 @@ wget https://paddleocr.bj.bcebos.com/dataset/XFUND.tar
```shell
python3.7 train_ser.py \
--model_name_or_path "layoutxlm-base-uncased" \
+ --ser_model_type "LayoutLM" \
--train_data_dir "XFUND/zh_train/image" \
--train_label_path "XFUND/zh_train/xfun_normalize_train.json" \
--eval_data_dir "XFUND/zh_val/image" \
@@ -155,6 +157,7 @@ python3.7 train_ser.py \
```shell
python3.7 train_ser.py \
--model_name_or_path "model_path" \
+ --ser_model_type "LayoutXLM" \
--train_data_dir "XFUND/zh_train/image" \
--train_label_path "XFUND/zh_train/xfun_normalize_train.json" \
--eval_data_dir "XFUND/zh_val/image" \
@@ -175,6 +178,7 @@ python3.7 train_ser.py \
export CUDA_VISIBLE_DEVICES=0
python3 eval_ser.py \
--model_name_or_path "PP-Layout_v1.0_ser_pretrained/" \
+ --ser_model_type "LayoutXLM" \
--eval_data_dir "XFUND/zh_val/image" \
--eval_label_path "XFUND/zh_val/xfun_normalize_val.json" \
--per_gpu_eval_batch_size 8 \
@@ -190,6 +194,7 @@ python3 eval_ser.py \
export CUDA_VISIBLE_DEVICES=0
python3.7 infer_ser.py \
--model_name_or_path "./PP-Layout_v1.0_ser_pretrained/" \
+ --ser_model_type "LayoutXLM" \
--output_dir "output_res/" \
--infer_imgs "XFUND/zh_val/image/" \
--ocr_json_path "XFUND/zh_val/xfun_normalize_val.json"
@@ -203,6 +208,7 @@ python3.7 infer_ser.py \
export CUDA_VISIBLE_DEVICES=0
python3.7 infer_ser_e2e.py \
--model_name_or_path "./output/PP-Layout_v1.0_ser_pretrained/" \
+ --ser_model_type "LayoutXLM" \
--max_seq_length 512 \
--output_dir "output_res_e2e/" \
--infer_imgs "images/input/zh_val_0.jpg"
diff --git a/ppstructure/vqa/eval_ser.py b/ppstructure/vqa/eval_ser.py
index acf37452a..52eeb8a1d 100644
--- a/ppstructure/vqa/eval_ser.py
+++ b/ppstructure/vqa/eval_ser.py
@@ -29,11 +29,21 @@ import paddle
import numpy as np
from seqeval.metrics import classification_report, f1_score, precision_score, recall_score
from paddlenlp.transformers import LayoutXLMModel, LayoutXLMTokenizer, LayoutXLMForTokenClassification
+from paddlenlp.transformers import LayoutLMModel, LayoutLMTokenizer, LayoutLMForTokenClassification
+
from xfun import XFUNDataset
+from losses import SERLoss
from utils import parse_args, get_bio_label_maps, print_arguments
from ppocr.utils.logging import get_logger
+MODELS = {
+ 'LayoutXLM':
+ (LayoutXLMTokenizer, LayoutXLMModel, LayoutXLMForTokenClassification),
+ 'LayoutLM':
+ (LayoutLMTokenizer, LayoutLMModel, LayoutLMForTokenClassification)
+}
+
def eval(args):
logger = get_logger()
@@ -42,9 +52,9 @@ def eval(args):
label2id_map, id2label_map = get_bio_label_maps(args.label_map_path)
pad_token_label_id = paddle.nn.CrossEntropyLoss().ignore_index
- tokenizer = LayoutXLMTokenizer.from_pretrained(args.model_name_or_path)
- model = LayoutXLMForTokenClassification.from_pretrained(
- args.model_name_or_path)
+ tokenizer_class, base_model_class, model_class = MODELS[args.ser_model_type]
+ tokenizer = tokenizer_class.from_pretrained(args.model_name_or_path)
+ model = model_class.from_pretrained(args.model_name_or_path)
eval_dataset = XFUNDataset(
tokenizer,
@@ -65,8 +75,11 @@ def eval(args):
use_shared_memory=True,
collate_fn=None, )
- results, _ = evaluate(args, model, tokenizer, eval_dataloader, label2id_map,
- id2label_map, pad_token_label_id, logger)
+ loss_class = SERLoss(len(label2id_map))
+
+ results, _ = evaluate(args, model, tokenizer, loss_class, eval_dataloader,
+ label2id_map, id2label_map, pad_token_label_id,
+ logger)
logger.info(results)
@@ -74,6 +87,7 @@ def eval(args):
def evaluate(args,
model,
tokenizer,
+ loss_class,
eval_dataloader,
label2id_map,
id2label_map,
@@ -88,24 +102,29 @@ def evaluate(args,
model.eval()
for idx, batch in enumerate(eval_dataloader):
with paddle.no_grad():
+ if args.ser_model_type == 'LayoutLM':
+ if 'image' in batch:
+ batch.pop('image')
+ labels = batch.pop('labels')
outputs = model(**batch)
- tmp_eval_loss, logits = outputs[:2]
+ if args.ser_model_type == 'LayoutXLM':
+ outputs = outputs[0]
+ loss = loss_class(labels, outputs, batch['attention_mask'])
- tmp_eval_loss = tmp_eval_loss.mean()
+ loss = loss.mean()
if paddle.distributed.get_rank() == 0:
logger.info("[Eval]process: {}/{}, loss: {:.5f}".format(
- idx, len(eval_dataloader), tmp_eval_loss.numpy()[0]))
+ idx, len(eval_dataloader), loss.numpy()[0]))
- eval_loss += tmp_eval_loss.item()
+ eval_loss += loss.item()
nb_eval_steps += 1
if preds is None:
- preds = logits.numpy()
- out_label_ids = batch["labels"].numpy()
+ preds = outputs.numpy()
+ out_label_ids = labels.numpy()
else:
- preds = np.append(preds, logits.numpy(), axis=0)
- out_label_ids = np.append(
- out_label_ids, batch["labels"].numpy(), axis=0)
+ preds = np.append(preds, outputs.numpy(), axis=0)
+ out_label_ids = np.append(out_label_ids, labels.numpy(), axis=0)
eval_loss = eval_loss / nb_eval_steps
preds = np.argmax(preds, axis=2)
diff --git a/ppstructure/vqa/infer_re.py b/ppstructure/vqa/infer_re.py
index 139ade9aa..2ffa60f5d 100644
--- a/ppstructure/vqa/infer_re.py
+++ b/ppstructure/vqa/infer_re.py
@@ -56,7 +56,11 @@ def infer(args):
ocr_info_list = load_ocr(args.eval_data_dir, args.eval_label_path)
for idx, batch in enumerate(eval_dataloader):
- logger.info("[Infer] process: {}/{}".format(idx, len(eval_dataloader)))
+ save_img_path = os.path.join(
+ args.output_dir,
+ os.path.splitext(os.path.basename(img_path))[0] + "_re.jpg")
+ logger.info("[Infer] process: {}/{}, save_result to {}".format(
+ idx, len(eval_dataloader), save_img_path))
with paddle.no_grad():
outputs = model(**batch)
pred_relations = outputs['pred_relations']
@@ -85,8 +89,7 @@ def infer(args):
img = cv2.imread(image_path)
img_show = draw_re_results(img, result)
- save_path = os.path.join(args.output_dir, os.path.basename(image_path))
- cv2.imwrite(save_path, img_show)
+ cv2.imwrite(save_img_path, img_show)
def load_ocr(img_folder, json_path):
diff --git a/ppstructure/vqa/infer_ser.py b/ppstructure/vqa/infer_ser.py
index 3c7ad03aa..2b715d7b9 100644
--- a/ppstructure/vqa/infer_ser.py
+++ b/ppstructure/vqa/infer_ser.py
@@ -24,6 +24,14 @@ import paddle
# relative reference
from utils import parse_args, get_image_file_list, draw_ser_results, get_bio_label_maps
from paddlenlp.transformers import LayoutXLMModel, LayoutXLMTokenizer, LayoutXLMForTokenClassification
+from paddlenlp.transformers import LayoutLMModel, LayoutLMTokenizer, LayoutLMForTokenClassification
+
+MODELS = {
+ 'LayoutXLM':
+ (LayoutXLMTokenizer, LayoutXLMModel, LayoutXLMForTokenClassification),
+ 'LayoutLM':
+ (LayoutLMTokenizer, LayoutLMModel, LayoutLMForTokenClassification)
+}
def pad_sentences(tokenizer,
@@ -217,10 +225,10 @@ def infer(args):
os.makedirs(args.output_dir, exist_ok=True)
# init token and model
- tokenizer = LayoutXLMTokenizer.from_pretrained(args.model_name_or_path)
- # model = LayoutXLMModel.from_pretrained(args.model_name_or_path)
- model = LayoutXLMForTokenClassification.from_pretrained(
- args.model_name_or_path)
+ tokenizer_class, base_model_class, model_class = MODELS[args.ser_model_type]
+ tokenizer = tokenizer_class.from_pretrained(args.model_name_or_path)
+ model = model_class.from_pretrained(args.model_name_or_path)
+
model.eval()
# load ocr results json
@@ -240,7 +248,10 @@ def infer(args):
"w",
encoding='utf-8') as fout:
for idx, img_path in enumerate(infer_imgs):
- print("process: [{}/{}]".format(idx, len(infer_imgs), img_path))
+ save_img_path = os.path.join(args.output_dir,
+ os.path.basename(img_path))
+ print("process: [{}/{}], save_result to {}".format(
+ idx, len(infer_imgs), save_img_path))
img = cv2.imread(img_path)
@@ -250,15 +261,21 @@ def infer(args):
ori_img=img,
ocr_info=ocr_info,
max_seq_len=args.max_seq_length)
+ if args.ser_model_type == 'LayoutLM':
+ preds = model(
+ input_ids=inputs["input_ids"],
+ bbox=inputs["bbox"],
+ token_type_ids=inputs["token_type_ids"],
+ attention_mask=inputs["attention_mask"])
+ elif args.ser_model_type == 'LayoutXLM':
+ preds = model(
+ input_ids=inputs["input_ids"],
+ bbox=inputs["bbox"],
+ image=inputs["image"],
+ token_type_ids=inputs["token_type_ids"],
+ attention_mask=inputs["attention_mask"])
+ preds = preds[0]
- outputs = model(
- input_ids=inputs["input_ids"],
- bbox=inputs["bbox"],
- image=inputs["image"],
- token_type_ids=inputs["token_type_ids"],
- attention_mask=inputs["attention_mask"])
-
- preds = outputs[0]
preds = postprocess(inputs["attention_mask"], preds,
args.label_map_path)
ocr_info = merge_preds_list_with_ocr_info(
@@ -271,9 +288,7 @@ def infer(args):
}, ensure_ascii=False) + "\n")
img_res = draw_ser_results(img, ocr_info)
- cv2.imwrite(
- os.path.join(args.output_dir, os.path.basename(img_path)),
- img_res)
+ cv2.imwrite(save_img_path, img_res)
return
diff --git a/ppstructure/vqa/infer_ser_e2e.py b/ppstructure/vqa/infer_ser_e2e.py
index a93a98ff6..05a029822 100644
--- a/ppstructure/vqa/infer_ser_e2e.py
+++ b/ppstructure/vqa/infer_ser_e2e.py
@@ -22,12 +22,20 @@ from PIL import Image
import paddle
from paddlenlp.transformers import LayoutXLMModel, LayoutXLMTokenizer, LayoutXLMForTokenClassification
+from paddlenlp.transformers import LayoutLMModel, LayoutLMTokenizer, LayoutLMForTokenClassification
# relative reference
from utils import parse_args, get_image_file_list, draw_ser_results, get_bio_label_maps
from utils import pad_sentences, split_page, preprocess, postprocess, merge_preds_list_with_ocr_info
+MODELS = {
+ 'LayoutXLM':
+ (LayoutXLMTokenizer, LayoutXLMModel, LayoutXLMForTokenClassification),
+ 'LayoutLM':
+ (LayoutLMTokenizer, LayoutLMModel, LayoutLMForTokenClassification)
+}
+
def trans_poly_to_bbox(poly):
x1 = np.min([p[0] for p in poly])
@@ -50,14 +58,15 @@ def parse_ocr_info_for_ser(ocr_result):
class SerPredictor(object):
def __init__(self, args):
-
+ self.args = args
self.max_seq_length = args.max_seq_length
# init ser token and model
- self.tokenizer = LayoutXLMTokenizer.from_pretrained(
- args.model_name_or_path)
- self.model = LayoutXLMForTokenClassification.from_pretrained(
+ tokenizer_class, base_model_class, model_class = MODELS[
+ args.ser_model_type]
+ self.tokenizer = tokenizer_class.from_pretrained(
args.model_name_or_path)
+ self.model = model_class.from_pretrained(args.model_name_or_path)
self.model.eval()
# init ocr_engine
@@ -89,14 +98,21 @@ class SerPredictor(object):
ocr_info=ocr_info,
max_seq_len=self.max_seq_length)
- outputs = self.model(
- input_ids=inputs["input_ids"],
- bbox=inputs["bbox"],
- image=inputs["image"],
- token_type_ids=inputs["token_type_ids"],
- attention_mask=inputs["attention_mask"])
+ if args.ser_model_type == 'LayoutLM':
+ preds = self.model(
+ input_ids=inputs["input_ids"],
+ bbox=inputs["bbox"],
+ token_type_ids=inputs["token_type_ids"],
+ attention_mask=inputs["attention_mask"])
+ elif args.ser_model_type == 'LayoutXLM':
+ preds = self.model(
+ input_ids=inputs["input_ids"],
+ bbox=inputs["bbox"],
+ image=inputs["image"],
+ token_type_ids=inputs["token_type_ids"],
+ attention_mask=inputs["attention_mask"])
+ preds = preds[0]
- preds = outputs[0]
preds = postprocess(inputs["attention_mask"], preds, self.id2label_map)
ocr_info = merge_preds_list_with_ocr_info(
ocr_info, inputs["segment_offset_id"], preds,
@@ -118,7 +134,11 @@ if __name__ == "__main__":
"w",
encoding='utf-8') as fout:
for idx, img_path in enumerate(infer_imgs):
- print("process: [{}/{}], {}".format(idx, len(infer_imgs), img_path))
+ save_img_path = os.path.join(
+ args.output_dir,
+ os.path.splitext(os.path.basename(img_path))[0] + "_ser.jpg")
+ print("process: [{}/{}], save_result to {}".format(
+ idx, len(infer_imgs), save_img_path))
img = cv2.imread(img_path)
@@ -129,7 +149,4 @@ if __name__ == "__main__":
}, ensure_ascii=False) + "\n")
img_res = draw_ser_results(img, result)
- cv2.imwrite(
- os.path.join(args.output_dir,
- os.path.splitext(os.path.basename(img_path))[0] +
- "_ser.jpg"), img_res)
+ cv2.imwrite(save_img_path, img_res)
diff --git a/ppstructure/vqa/losses.py b/ppstructure/vqa/losses.py
new file mode 100644
index 000000000..e8dad01c3
--- /dev/null
+++ b/ppstructure/vqa/losses.py
@@ -0,0 +1,35 @@
+# Copyright (c) 2021 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.
+
+from paddle import nn
+
+
+class SERLoss(nn.Layer):
+ def __init__(self, num_classes):
+ super().__init__()
+ self.loss_class = nn.CrossEntropyLoss()
+ self.num_classes = num_classes
+ self.ignore_index = self.loss_class.ignore_index
+
+ def forward(self, labels, outputs, attention_mask):
+ if attention_mask is not None:
+ active_loss = attention_mask.reshape([-1, ]) == 1
+ active_outputs = outputs.reshape(
+ [-1, self.num_classes])[active_loss]
+ active_labels = labels.reshape([-1, ])[active_loss]
+ loss = self.loss_class(active_outputs, active_labels)
+ else:
+ loss = self.loss_class(
+ outputs.reshape([-1, self.num_classes]), labels.reshape([-1, ]))
+ return loss
diff --git a/ppstructure/vqa/train_ser.py b/ppstructure/vqa/train_ser.py
index 6791cea89..2670ef9ee 100644
--- a/ppstructure/vqa/train_ser.py
+++ b/ppstructure/vqa/train_ser.py
@@ -29,11 +29,21 @@ import paddle
import numpy as np
from seqeval.metrics import classification_report, f1_score, precision_score, recall_score
from paddlenlp.transformers import LayoutXLMModel, LayoutXLMTokenizer, LayoutXLMForTokenClassification
+from paddlenlp.transformers import LayoutLMModel, LayoutLMTokenizer, LayoutLMForTokenClassification
+
from xfun import XFUNDataset
from utils import parse_args, get_bio_label_maps, print_arguments, set_seed
from eval_ser import evaluate
+from losses import SERLoss
from ppocr.utils.logging import get_logger
+MODELS = {
+ 'LayoutXLM':
+ (LayoutXLMTokenizer, LayoutXLMModel, LayoutXLMForTokenClassification),
+ 'LayoutLM':
+ (LayoutLMTokenizer, LayoutLMModel, LayoutLMForTokenClassification)
+}
+
def train(args):
os.makedirs(args.output_dir, exist_ok=True)
@@ -44,22 +54,24 @@ def train(args):
print_arguments(args, logger)
label2id_map, id2label_map = get_bio_label_maps(args.label_map_path)
- pad_token_label_id = paddle.nn.CrossEntropyLoss().ignore_index
+ loss_class = SERLoss(len(label2id_map))
+
+ pad_token_label_id = loss_class.ignore_index
# dist mode
if distributed:
paddle.distributed.init_parallel_env()
- tokenizer = LayoutXLMTokenizer.from_pretrained(args.model_name_or_path)
+ tokenizer_class, base_model_class, model_class = MODELS[args.ser_model_type]
+ tokenizer = tokenizer_class.from_pretrained(args.model_name_or_path)
if not args.resume:
- model = LayoutXLMModel.from_pretrained(args.model_name_or_path)
- model = LayoutXLMForTokenClassification(
- model, num_classes=len(label2id_map), dropout=None)
+ base_model = base_model_class.from_pretrained(args.model_name_or_path)
+ model = model_class(
+ base_model, num_classes=len(label2id_map), dropout=None)
logger.info('train from scratch')
else:
logger.info('resume from {}'.format(args.model_name_or_path))
- model = LayoutXLMForTokenClassification.from_pretrained(
- args.model_name_or_path)
+ model = model_class.from_pretrained(args.model_name_or_path)
# dist mode
if distributed:
@@ -153,12 +165,19 @@ def train(args):
for step, batch in enumerate(train_dataloader):
train_reader_cost += time.time() - reader_start
+ if args.ser_model_type == 'LayoutLM':
+ if 'image' in batch:
+ batch.pop('image')
+ labels = batch.pop('labels')
+
train_start = time.time()
outputs = model(**batch)
train_run_cost += time.time() - train_start
+ if args.ser_model_type == 'LayoutXLM':
+ outputs = outputs[0]
+ loss = loss_class(labels, outputs, batch['attention_mask'])
# model outputs are always tuple in ppnlp (see doc)
- loss = outputs[0]
loss = loss.mean()
loss.backward()
tr_loss += loss.item()
@@ -166,7 +185,7 @@ def train(args):
lr_scheduler.step() # Update learning rate schedule
optimizer.clear_grad()
global_step += 1
- total_samples += batch['image'].shape[0]
+ total_samples += batch['input_ids'].shape[0]
if rank == 0 and step % print_step == 0:
logger.info(
@@ -186,9 +205,9 @@ def train(args):
if rank == 0 and args.eval_steps > 0 and global_step % args.eval_steps == 0 and args.evaluate_during_training:
# Log metrics
# Only evaluate when single GPU otherwise metrics may not average well
- results, _ = evaluate(args, model, tokenizer, eval_dataloader,
- label2id_map, id2label_map,
- pad_token_label_id, logger)
+ results, _ = evaluate(args, model, tokenizer, loss_class,
+ eval_dataloader, label2id_map,
+ id2label_map, pad_token_label_id, logger)
if best_metrics is None or results["f1"] >= best_metrics["f1"]:
best_metrics = copy.deepcopy(results)
@@ -201,7 +220,8 @@ def train(args):
tokenizer.save_pretrained(output_dir)
paddle.save(args,
os.path.join(output_dir, "training_args.bin"))
- logger.info("Saving model checkpoint to %s", output_dir)
+ logger.info("Saving model checkpoint to {}".format(
+ output_dir))
logger.info("[epoch {}/{}][iter: {}/{}] results: {}".format(
epoch_id, args.num_train_epochs, step,
@@ -219,7 +239,7 @@ def train(args):
model.save_pretrained(output_dir)
tokenizer.save_pretrained(output_dir)
paddle.save(args, os.path.join(output_dir, "training_args.bin"))
- logger.info("Saving model checkpoint to %s", output_dir)
+ logger.info("Saving model checkpoint to {}".format(output_dir))
return global_step, tr_loss / global_step
diff --git a/ppstructure/vqa/utils.py b/ppstructure/vqa/utils.py
index 414317c0d..b9f2edc86 100644
--- a/ppstructure/vqa/utils.py
+++ b/ppstructure/vqa/utils.py
@@ -350,6 +350,8 @@ def parse_args():
# yapf: disable
parser.add_argument("--model_name_or_path",
default=None, type=str, required=True,)
+ parser.add_argument("--ser_model_type",
+ default='LayoutXLM', type=str)
parser.add_argument("--re_model_name_or_path",
default=None, type=str, required=False,)
parser.add_argument("--train_data_dir", default=None,
From 9131c4a7acc12ba79c7a60aef72a94abb5d730eb Mon Sep 17 00:00:00 2001
From: WenmuZhou <572459439@qq.com>
Date: Mon, 20 Dec 2021 14:38:49 +0000
Subject: [PATCH 02/35] add LayoutLM ser
---
ppstructure/vqa/README.md | 10 ++---
ppstructure/vqa/infer.sh | 61 +++++++++++++++++++++++++++++
ppstructure/vqa/infer_re.py | 10 ++---
ppstructure/vqa/infer_ser_e2e.py | 4 +-
ppstructure/vqa/infer_ser_re_e2e.py | 11 +++---
5 files changed, 79 insertions(+), 17 deletions(-)
create mode 100644 ppstructure/vqa/infer.sh
diff --git a/ppstructure/vqa/README.md b/ppstructure/vqa/README.md
index 51a616c43..975139c79 100644
--- a/ppstructure/vqa/README.md
+++ b/ppstructure/vqa/README.md
@@ -195,7 +195,7 @@ export CUDA_VISIBLE_DEVICES=0
python3.7 infer_ser.py \
--model_name_or_path "./PP-Layout_v1.0_ser_pretrained/" \
--ser_model_type "LayoutXLM" \
- --output_dir "output_res/" \
+ --output_dir "output/ser/" \
--infer_imgs "XFUND/zh_val/image/" \
--ocr_json_path "XFUND/zh_val/xfun_normalize_val.json"
```
@@ -210,7 +210,7 @@ python3.7 infer_ser_e2e.py \
--model_name_or_path "./output/PP-Layout_v1.0_ser_pretrained/" \
--ser_model_type "LayoutXLM" \
--max_seq_length 512 \
- --output_dir "output_res_e2e/" \
+ --output_dir "output/ser_e2e/" \
--infer_imgs "images/input/zh_val_0.jpg"
```
@@ -284,7 +284,7 @@ python3 eval_re.py \
--eval_data_dir "XFUND/zh_val/image" \
--eval_label_path "XFUND/zh_val/xfun_normalize_val.json" \
--label_map_path 'labels/labels_ser.txt' \
- --output_dir "output/re_test/" \
+ --output_dir "output/re/" \
--per_gpu_eval_batch_size 8 \
--num_workers 8 \
--seed 2048
@@ -302,7 +302,7 @@ python3 infer_re.py \
--eval_data_dir "XFUND/zh_val/image" \
--eval_label_path "XFUND/zh_val/xfun_normalize_val.json" \
--label_map_path 'labels/labels_ser.txt' \
- --output_dir "output_res" \
+ --output_dir "output/re/" \
--per_gpu_eval_batch_size 1 \
--seed 2048
```
@@ -317,7 +317,7 @@ python3.7 infer_ser_re_e2e.py \
--model_name_or_path "./PP-Layout_v1.0_ser_pretrained/" \
--re_model_name_or_path "./PP-Layout_v1.0_re_pretrained/" \
--max_seq_length 512 \
- --output_dir "output_ser_re_e2e_train/" \
+ --output_dir "output/ser_re_e2e/" \
--infer_imgs "images/input/zh_val_21.jpg"
```
diff --git a/ppstructure/vqa/infer.sh b/ppstructure/vqa/infer.sh
new file mode 100644
index 000000000..2cd1cea44
--- /dev/null
+++ b/ppstructure/vqa/infer.sh
@@ -0,0 +1,61 @@
+export CUDA_VISIBLE_DEVICES=6
+# python3.7 infer_ser_e2e.py \
+# --model_name_or_path "output/ser_distributed/best_model" \
+# --max_seq_length 512 \
+# --output_dir "output_res_e2e/" \
+# --infer_imgs "/ssd1/zhoujun20/VQA/data/XFUN_v1.0_data/zh.val/zh_val_0.jpg"
+
+
+# python3.7 infer_ser_re_e2e.py \
+# --model_name_or_path "output/ser_distributed/best_model" \
+# --re_model_name_or_path "output/re_test/best_model" \
+# --max_seq_length 512 \
+# --output_dir "output_ser_re_e2e_train/" \
+# --infer_imgs "images/input/zh_val_21.jpg"
+
+# python3.7 infer_ser.py \
+# --model_name_or_path "output/ser_LayoutLM/best_model" \
+# --ser_model_type "LayoutLM" \
+# --output_dir "ser_LayoutLM/" \
+# --infer_imgs "images/input/zh_val_21.jpg" \
+# --ocr_json_path "/ssd1/zhoujun20/VQA/data/XFUN_v1.0_data/xfun_normalize_val.json"
+
+python3.7 infer_ser.py \
+ --model_name_or_path "output/ser_new/best_model" \
+ --ser_model_type "LayoutXLM" \
+ --output_dir "ser_new/" \
+ --infer_imgs "images/input/zh_val_21.jpg" \
+ --ocr_json_path "/ssd1/zhoujun20/VQA/data/XFUN_v1.0_data/xfun_normalize_val.json"
+
+# python3.7 infer_ser_e2e.py \
+# --model_name_or_path "output/ser_new/best_model" \
+# --ser_model_type "LayoutXLM" \
+# --max_seq_length 512 \
+# --output_dir "output/ser_new/" \
+# --infer_imgs "images/input/zh_val_0.jpg"
+
+
+# python3.7 infer_ser_e2e.py \
+# --model_name_or_path "output/ser_LayoutLM/best_model" \
+# --ser_model_type "LayoutLM" \
+# --max_seq_length 512 \
+# --output_dir "output/ser_LayoutLM/" \
+# --infer_imgs "images/input/zh_val_0.jpg"
+
+# python3 infer_re.py \
+# --model_name_or_path "/ssd1/zhoujun20/VQA/PaddleOCR/ppstructure/vqa/output/re_test/best_model/" \
+# --max_seq_length 512 \
+# --eval_data_dir "/ssd1/zhoujun20/VQA/data/XFUN_v1.0_data/zh.val" \
+# --eval_label_path "/ssd1/zhoujun20/VQA/data/XFUN_v1.0_data/xfun_normalize_val.json" \
+# --label_map_path 'labels/labels_ser.txt' \
+# --output_dir "output_res" \
+# --per_gpu_eval_batch_size 1 \
+# --seed 2048
+
+# python3.7 infer_ser_re_e2e.py \
+# --model_name_or_path "output/ser_LayoutLM/best_model" \
+# --ser_model_type "LayoutLM" \
+# --re_model_name_or_path "output/re_new/best_model" \
+# --max_seq_length 512 \
+# --output_dir "output_ser_re_e2e/" \
+# --infer_imgs "images/input/zh_val_21.jpg"
\ No newline at end of file
diff --git a/ppstructure/vqa/infer_re.py b/ppstructure/vqa/infer_re.py
index 2ffa60f5d..98c61bacc 100644
--- a/ppstructure/vqa/infer_re.py
+++ b/ppstructure/vqa/infer_re.py
@@ -56,19 +56,19 @@ def infer(args):
ocr_info_list = load_ocr(args.eval_data_dir, args.eval_label_path)
for idx, batch in enumerate(eval_dataloader):
+ ocr_info = ocr_info_list[idx]
+ image_path = ocr_info['image_path']
+ ocr_info = ocr_info['ocr_info']
+
save_img_path = os.path.join(
args.output_dir,
- os.path.splitext(os.path.basename(img_path))[0] + "_re.jpg")
+ os.path.splitext(os.path.basename(image_path))[0] + "_re.jpg")
logger.info("[Infer] process: {}/{}, save_result to {}".format(
idx, len(eval_dataloader), save_img_path))
with paddle.no_grad():
outputs = model(**batch)
pred_relations = outputs['pred_relations']
- ocr_info = ocr_info_list[idx]
- image_path = ocr_info['image_path']
- ocr_info = ocr_info['ocr_info']
-
# 根据entity里的信息，做token解码后去过滤不要的ocr_info
ocr_info = filter_bg_by_txt(ocr_info, batch, tokenizer)
diff --git a/ppstructure/vqa/infer_ser_e2e.py b/ppstructure/vqa/infer_ser_e2e.py
index 05a029822..bceb3434b 100644
--- a/ppstructure/vqa/infer_ser_e2e.py
+++ b/ppstructure/vqa/infer_ser_e2e.py
@@ -98,13 +98,13 @@ class SerPredictor(object):
ocr_info=ocr_info,
max_seq_len=self.max_seq_length)
- if args.ser_model_type == 'LayoutLM':
+ if self.args.ser_model_type == 'LayoutLM':
preds = self.model(
input_ids=inputs["input_ids"],
bbox=inputs["bbox"],
token_type_ids=inputs["token_type_ids"],
attention_mask=inputs["attention_mask"])
- elif args.ser_model_type == 'LayoutXLM':
+ elif self.args.ser_model_type == 'LayoutXLM':
preds = self.model(
input_ids=inputs["input_ids"],
bbox=inputs["bbox"],
diff --git a/ppstructure/vqa/infer_ser_re_e2e.py b/ppstructure/vqa/infer_ser_re_e2e.py
index 23737406d..a6316b625 100644
--- a/ppstructure/vqa/infer_ser_re_e2e.py
+++ b/ppstructure/vqa/infer_ser_re_e2e.py
@@ -117,7 +117,11 @@ if __name__ == "__main__":
"w",
encoding='utf-8') as fout:
for idx, img_path in enumerate(infer_imgs):
- print("process: [{}/{}], {}".format(idx, len(infer_imgs), img_path))
+ save_img_path = os.path.join(
+ args.output_dir,
+ os.path.splitext(os.path.basename(img_path))[0] + "_re.jpg")
+ print("process: [{}/{}], save_result to {}".format(
+ idx, len(infer_imgs), save_img_path))
img = cv2.imread(img_path)
@@ -128,7 +132,4 @@ if __name__ == "__main__":
}, ensure_ascii=False) + "\n")
img_res = draw_re_results(img, result)
- cv2.imwrite(
- os.path.join(args.output_dir,
- os.path.splitext(os.path.basename(img_path))[0] +
- "_re.jpg"), img_res)
+ cv2.imwrite(save_img_path, img_res)
From 11d6814f365b2e881adbdedc9865f86dae3022a9 Mon Sep 17 00:00:00 2001
From: WenmuZhou <572459439@qq.com>
Date: Mon, 20 Dec 2021 14:47:43 +0000
Subject: [PATCH 03/35] rm _
---
ppstructure/vqa/infer_re.py | 2 +-
ppstructure/vqa/infer_ser.py | 2 +-
ppstructure/vqa/infer_ser_e2e.py | 2 +-
ppstructure/vqa/infer_ser_re_e2e.py | 2 +-
4 files changed, 4 insertions(+), 4 deletions(-)
diff --git a/ppstructure/vqa/infer_re.py b/ppstructure/vqa/infer_re.py
index 98c61bacc..7937700a7 100644
--- a/ppstructure/vqa/infer_re.py
+++ b/ppstructure/vqa/infer_re.py
@@ -63,7 +63,7 @@ def infer(args):
save_img_path = os.path.join(
args.output_dir,
os.path.splitext(os.path.basename(image_path))[0] + "_re.jpg")
- logger.info("[Infer] process: {}/{}, save_result to {}".format(
+ logger.info("[Infer] process: {}/{}, save result to {}".format(
idx, len(eval_dataloader), save_img_path))
with paddle.no_grad():
outputs = model(**batch)
diff --git a/ppstructure/vqa/infer_ser.py b/ppstructure/vqa/infer_ser.py
index 2b715d7b9..7994b5449 100644
--- a/ppstructure/vqa/infer_ser.py
+++ b/ppstructure/vqa/infer_ser.py
@@ -250,7 +250,7 @@ def infer(args):
for idx, img_path in enumerate(infer_imgs):
save_img_path = os.path.join(args.output_dir,
os.path.basename(img_path))
- print("process: [{}/{}], save_result to {}".format(
+ print("process: [{}/{}], save result to {}".format(
idx, len(infer_imgs), save_img_path))
img = cv2.imread(img_path)
diff --git a/ppstructure/vqa/infer_ser_e2e.py b/ppstructure/vqa/infer_ser_e2e.py
index bceb3434b..6bb024750 100644
--- a/ppstructure/vqa/infer_ser_e2e.py
+++ b/ppstructure/vqa/infer_ser_e2e.py
@@ -137,7 +137,7 @@ if __name__ == "__main__":
save_img_path = os.path.join(
args.output_dir,
os.path.splitext(os.path.basename(img_path))[0] + "_ser.jpg")
- print("process: [{}/{}], save_result to {}".format(
+ print("process: [{}/{}], save result to {}".format(
idx, len(infer_imgs), save_img_path))
img = cv2.imread(img_path)
diff --git a/ppstructure/vqa/infer_ser_re_e2e.py b/ppstructure/vqa/infer_ser_re_e2e.py
index a6316b625..32d8850a1 100644
--- a/ppstructure/vqa/infer_ser_re_e2e.py
+++ b/ppstructure/vqa/infer_ser_re_e2e.py
@@ -120,7 +120,7 @@ if __name__ == "__main__":
save_img_path = os.path.join(
args.output_dir,
os.path.splitext(os.path.basename(img_path))[0] + "_re.jpg")
- print("process: [{}/{}], save_result to {}".format(
+ print("process: [{}/{}], save result to {}".format(
idx, len(infer_imgs), save_img_path))
img = cv2.imread(img_path)
From bbcabd8145306ea84e1489202c88315981de1ed1 Mon Sep 17 00:00:00 2001
From: changy1105 <18392600593@163.com>
Date: Mon, 20 Dec 2021 22:54:41 +0800
Subject: [PATCH 04/35] [TIPC] Add js infer test
---
.gitignore | 2 +
test_tipc/docs/test_inference_js.md | 50 +++++++++++
test_tipc/prepare_js.sh | 92 ++++++++++++++++++++
test_tipc/test_inference_js.sh | 8 ++
test_tipc/web/expect.json | 20 +++++
test_tipc/web/index.html | 13 +++
test_tipc/web/index.test.js | 82 ++++++++++++++++++
test_tipc/web/jest-puppeteer.config.js | 14 ++++
test_tipc/web/jest.config.js | 111 +++++++++++++++++++++++++
test_tipc/web/test.jpg | Bin 0 -> 287898 bytes
10 files changed, 392 insertions(+)
create mode 100644 test_tipc/docs/test_inference_js.md
create mode 100644 test_tipc/prepare_js.sh
create mode 100644 test_tipc/test_inference_js.sh
create mode 100644 test_tipc/web/expect.json
create mode 100644 test_tipc/web/index.html
create mode 100644 test_tipc/web/index.test.js
create mode 100644 test_tipc/web/jest-puppeteer.config.js
create mode 100644 test_tipc/web/jest.config.js
create mode 100644 test_tipc/web/test.jpg
diff --git a/.gitignore b/.gitignore
index 9d85e7a8c..caf886a2b 100644
--- a/.gitignore
+++ b/.gitignore
@@ -29,3 +29,5 @@ paddleocr.egg-info/
/deploy/android_demo/app/PaddleLite/
/deploy/android_demo/app/.cxx/
/deploy/android_demo/app/cache/
+test_tipc/web/models/
+test_tipc/web/node_modules/
diff --git a/test_tipc/docs/test_inference_js.md b/test_tipc/docs/test_inference_js.md
new file mode 100644
index 000000000..c0b7d653a
--- /dev/null
+++ b/test_tipc/docs/test_inference_js.md
@@ -0,0 +1,50 @@
+# Web 端基础预测功能测试
+
+Web 端主要基于 Jest-Puppeteer 完成 e2e 测试，其中 Puppeteer 操作 Chrome 完成推理流程，Jest 完成测试流程。
+>Puppeteer 是一个 Node 库，它提供了一个高级 API 来通过 DevTools 协议控制 Chromium 或 Chrome
+>Jest 是一个 JavaScript 测试框架，旨在确保任何 JavaScript 代码的正确性。
+#### 环境准备
+
+* 安装 Node（包含 npm ） （https://nodejs.org/zh-cn/download/）
+* 确认是否安装成功，在命令行执行
+```sh
+# 显示所安 node 版本号，即表示成功安装
+node -v
+```
+* 确认 npm 是否安装成成
+```sh
+# npm 随着 node 一起安装，一般无需额外安装
+# 显示所安 npm 版本号，即表示成功安装
+npm -v
+```
+
+#### 使用
+```sh
+# web 测试环境准备
+bash test_tipc/prepare_js.sh 'js_infer'
+# web 推理测试
+bash test_tipc/test_inference_js.sh
+```
+
+#### 流程设计
+
+###### paddlejs prepare
+ 1. 判断 node, npm 是否安装
+ 2. 下载测试模型，当前检测模型是 ch_PP-OCRv2_det_infer ，识别模型是 ch_PP-OCRv2_rec_infer[1, 3, 32, 320]。如果需要替换模型，可直接将模型文件放在test_tipc/web/models/目录下。
+ - 文本检测模型：https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_infer.tar
+ - 文本识别模型：https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_infer.tar
+ - 文本识别模型[1, 3, 32, 320]：https://paddlejs.bj.bcebos.com/models/ch_PP-OCRv2_rec_infer.tar
+ - 保证较为准确的识别效果，需要将文本识别模型导出为输入shape是[1, 3, 32, 320]的静态模型
+ 3. 转换模型， model.pdmodel model.pdiparams 转换为 model.json chunk.dat（检测模型保存地址：test_tipc/web/models/ch_PP-OCRv2/det，识别模型保存地址：test_tipc/web/models/ch_PP-OCRv2/rec）
+ 4. 安装最新版本 ocr sdk @paddlejs-models/ocr@latest
+ 5. 安装测试环境依赖 puppeteer、jest、jest-puppeteer，如果检查到已经安装，则不会进行二次安装
+
+ ###### paddlejs infer test
+ 1. Jest 执行 server command：`python3 -m http.server 9811` 开启本地服务
+ 2. 启动 Jest 测试服务，通过 jest-puppeteer 插件完成 chrome 操作，加载 @paddlejs-models/ocr 脚本完成推理流程
+ 3. 测试用例为原图识别后的文本结果与预期文本结果（expect.json）进行对比，测试通过有两个标准：
+ * 原图识别结果逐字符与预期结果对比，误差不超过 **10个字符**；
+ * 原图识别结果每个文本框字符内容与预期结果进行相似度对比，相似度不小于 0.9（全部一致则相似度为1）。
+
+ 只有满足上述两个标准，视为测试通过。通过为如下显示：
+ 
diff --git a/test_tipc/prepare_js.sh b/test_tipc/prepare_js.sh
new file mode 100644
index 000000000..7bcdeb35a
--- /dev/null
+++ b/test_tipc/prepare_js.sh
@@ -0,0 +1,92 @@
+#!/bin/bash
+
+set -o errexit
+set -o nounset
+shopt -s extglob
+
+# paddlejs prepare 主要流程
+# 1. 判断 node, npm 是否安装
+# 2. 下载测试模型，当前检测模型是 ch_PP-OCRv2_det_infer ，识别模型是 ch_PP-OCRv2_rec_infer [1, 3, 32, 320]。如果需要替换模型，可直接将模型文件放在test_tipc/web/models/目录下。
+# - 文本检测模型：https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_infer.tar
+# - 文本识别模型：https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_infer.tar
+# - 文本识别模型[1, 3, 32, 320]：https://paddlejs.bj.bcebos.com/models/ch_PP-OCRv2_rec_infer.tar
+# - 保证较为准确的识别效果，需要将文本识别模型导出为输入shape[1, 3, 32, 320]的静态模型
+# 3. 转换模型， model.pdmodel model.pdiparams 转换为 model.json chunk.dat（检测模型保存地址：test_tipc/web/models/ch_PP-OCRv2/det，识别模型保存地址：test_tipc/web/models/ch_PP-OCRv2/rec）
+# 4. 安装最新版本 ocr sdk @paddlejs-models/ocr@latest
+# 5. 安装测试环境依赖 puppeteer、jest、jest-puppeteer，如果检查到已经安装，则不会进行二次安装
+
+# 判断是否安装了node
+if ! type node >/dev/null 2>&1; then
+ echo -e "\033[31m node 未安装 \033[0m"
+ exit
+fi
+
+# 判断是否安装了npm
+if ! type npm >/dev/null 2>&1; then
+ echo -e "\033[31m npm 未安装 \033[0m"
+ exit
+fi
+
+# MODE be 'js_infer'
+MODE=$1
+# js_infer MODE , load model file and convert model to js_infer
+if [ ${MODE} != "js_infer" ];then
+ echo "Please change mode to 'js_infer'"
+ exit
+fi
+
+
+# saved_model_name
+det_saved_model_name=ch_PP-OCRv2_det_infer
+rec_saved_model_name=ch_PP-OCRv2_rec_infer
+
+# model_path
+model_path=test_tipc/web/models/
+
+rm -rf $model_path
+
+echo ${model_path}${det_saved_model_name}
+echo ${model_path}${rec_saved_model_name}
+
+# download ocr_det inference model
+wget -nc -P $model_path https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_infer.tar
+cd $model_path && tar xf ch_PP-OCRv2_det_infer.tar && cd ../../../
+
+# download ocr_rec inference model
+wget -nc -P $model_path https://paddlejs.bj.bcebos.com/models/ch_PP-OCRv2_rec_infer.tar
+cd $model_path && tar xf ch_PP-OCRv2_rec_infer.tar && cd ../../../
+
+MYDIR=`pwd`
+echo $MYDIR
+
+pip3 install paddlejsconverter
+
+# convert inference model to web model: model.json、chunk.dat
+paddlejsconverter \
+ --modelPath=$model_path$det_saved_model_name/inference.pdmodel \
+ --paramPath=$model_path$det_saved_model_name/inference.pdiparams \
+ --outputDir=$model_path$det_saved_model_name/ \
+
+paddlejsconverter \
+ --modelPath=$model_path$rec_saved_model_name/inference.pdmodel \
+ --paramPath=$model_path$rec_saved_model_name/inference.pdiparams \
+ --outputDir=$model_path$rec_saved_model_name/ \
+
+# always install latest ocr sdk
+cd test_tipc/web
+echo -e "\033[33m Installing the latest ocr sdk... \033[0m"
+npm install @paddlejs-models/ocr@latest
+npm info @paddlejs-models/ocr
+echo -e "\033[32m The latest ocr sdk installed completely.!~ \033[0m"
+
+# install dependencies
+if [ `npm list --dept 0 | grep puppeteer | wc -l` -ne 0 ] && [ `npm list --dept 0 | grep jest | wc -l` -ne 0 ];then
+ echo -e "\033[32m Dependencies have installed \033[0m"
+else
+ echo -e "\033[33m Installing dependencies ... \033[0m"
+ npm install jest jest-puppeteer puppeteer
+ echo -e "\033[32m Dependencies installed completely.!~ \033[0m"
+fi
+
+# del package-lock.json
+rm package-lock.json
diff --git a/test_tipc/test_inference_js.sh b/test_tipc/test_inference_js.sh
new file mode 100644
index 000000000..e96b18752
--- /dev/null
+++ b/test_tipc/test_inference_js.sh
@@ -0,0 +1,8 @@
+#!/bin/bash
+
+set -o errexit
+set -o nounset
+
+cd test_tipc/web
+# run ocr test in chrome
+./node_modules/.bin/jest --config ./jest.config.js
diff --git a/test_tipc/web/expect.json b/test_tipc/web/expect.json
new file mode 100644
index 000000000..a60c80a7c
--- /dev/null
+++ b/test_tipc/web/expect.json
@@ -0,0 +1,20 @@
+{
+ "text": [
+ "纯臻营养护发素",
+ "产品信息/参数",
+ "（45元/每公斤，100公斤起订）",
+ "每瓶22元，1000瓶起订）",
+ "【品牌】：代加工方式/OEMODM",
+ "【品名】：纯臻营养护发素",
+ "【产品编号】：YM-X-3011",
+ "ODMOEM",
+ "【净含量】：220ml",
+ "【适用人群】：适合所有肤质",
+ "【主要成分】：鲸蜡硬脂醇、燕麦β-葡聚",
+ "糖、椰油酰胺丙基甜菜碱、泛醌",
+ "（成品包材）",
+ "【主要功能】：可紧致头发磷层，从而达到",
+ "即时持久改善头发光泽的效果，给干燥的头",
+ "发足够的滋养"
+ ]
+}
diff --git a/test_tipc/web/index.html b/test_tipc/web/index.html
new file mode 100644
index 000000000..39921fbf0
--- /dev/null
+++ b/test_tipc/web/index.html
@@ -0,0 +1,13 @@
+
+
+
+
+
+
+ ocr test
+
+
+ 
+
+
+
\ No newline at end of file
diff --git a/test_tipc/web/index.test.js b/test_tipc/web/index.test.js
new file mode 100644
index 000000000..e07aed82e
--- /dev/null
+++ b/test_tipc/web/index.test.js
@@ -0,0 +1,82 @@
+const expectData = require('./expect.json');
+
+describe('e2e test ocr model', () => {
+
+ beforeAll(async () => {
+ await page.goto(PATH);
+ });
+
+ it('ocr infer and diff test', async () => {
+ page.on('console', msg => console.log('PAGE LOG:', msg.text()));
+
+ const text = await page.evaluate(async () => {
+ const $ocr = document.querySelector('#ocr');
+ const ocr = paddlejs['ocr'];
+ await ocr.init('./models/ch_PP-OCRv2_det_infer', './models/ch_PP-OCRv2_rec_infer');
+ const res = await ocr.recognize($ocr);
+ return res.text;
+ });
+ // 模型文字识别结果与预期结果diff的字符数
+ let diffNum = 0;
+ // 文本框字符串相似度
+ let similarity = 0;
+ // 预期字符diff数
+ const expectedDiffNum = 10;
+ // 预期文本框字符串相似度
+ const expecteSimilarity = 0.9;
+ // 预期文本内容
+ const expectResult = expectData.text;
+
+ expectResult && expectResult.forEach((item, index) => {
+ const word = text[index];
+ // 逐字符对比
+ for(let i = 0; i < item.length; i++) {
+ if (item[i] !== word[i]) {
+ console.log('expect: ', item[i], ' word: ', word[i]);
+ diffNum++;
+ }
+ }
+ // 文本框字符串相似度对比
+ const s = similar(item, word);
+ similarity += s;
+ });
+
+ similarity = similarity / expectResult.length;
+
+ expect(diffNum).toBeLessThanOrEqual(expectedDiffNum);
+
+ expect(similarity).toBeGreaterThanOrEqual(expecteSimilarity);
+
+ function similar(string, expect) {
+ if (!string || !expect) {
+ return 0;
+ }
+ const length = string.length > expect.length ? string.length : expect.length;
+ const n = string.length;
+ const m = expect.length;
+ let data = [];
+ const min = (a, b, c) => {
+ return a < b ? (a < c ? a : c) : (b < c ? b : c);
+ };
+ let i, j, si, ej, cost;
+ if (n === 0) return m;
+ if (m === 0) return n;
+ for (i = 0; i <= n; i++) {
+ data[i] = [];
+ [i][0] = i
+ }
+ for (j = 0; j <= m; j++) {
+ data[0][j] = j;
+ }
+ for (i = 1; i <= n; i++) {
+ si = string.charAt(i - 1);
+ for (j = 1; j <= m; j++) {
+ ej = expect.charAt(j - 1);
+ cost = si === ej ? 0 : 1;
+ data[i][j] = min(data[i - 1][j] + 1, data[i][j - 1] + 1, data[i - 1][j - 1] + cost);
+ }
+ }
+ return (1 - data[n][m] / length);
+ }
+ });
+});
diff --git a/test_tipc/web/jest-puppeteer.config.js b/test_tipc/web/jest-puppeteer.config.js
new file mode 100644
index 000000000..ac60eea6b
--- /dev/null
+++ b/test_tipc/web/jest-puppeteer.config.js
@@ -0,0 +1,14 @@
+// jest-puppeteer.config.js
+module.exports = {
+ launch: {
+ headless: false,
+ product: 'chrome'
+ },
+ browserContext: 'default',
+ server: {
+ command: 'python3 -m http.server 9811',
+ port: 9811,
+ launchTimeout: 10000,
+ debug: true
+ }
+};
diff --git a/test_tipc/web/jest.config.js b/test_tipc/web/jest.config.js
new file mode 100644
index 000000000..aed1573ef
--- /dev/null
+++ b/test_tipc/web/jest.config.js
@@ -0,0 +1,111 @@
+// For a detailed explanation regarding each configuration property and type check, visit:
+// https://jestjs.io/docs/en/configuration.html
+
+module.exports = {
+ preset: 'jest-puppeteer',
+ // All imported modules in your tests should be mocked automatically
+ // automock: false,
+
+ // Automatically clear mock calls and instances between every test
+ clearMocks: true,
+
+ // An object that configures minimum threshold enforcement for coverage results
+ // coverageThreshold: undefined,
+
+ // A set of global variables that need to be available in all test environments
+ globals: {
+ PATH: 'http://localhost:9811'
+ },
+
+ // The maximum amount of workers used to run your tests. Can be specified as % or a number. E.g. maxWorkers: 10% will use 10% of your CPU amount + 1 as the maximum worker number. maxWorkers: 2 will use a maximum of 2 workers.
+ // maxWorkers: "50%",
+
+ // An array of directory names to be searched recursively up from the requiring module's location
+ // moduleDirectories: [
+ // "node_modules"
+ // ],
+
+ // An array of file extensions your modules use
+ moduleFileExtensions: [
+ 'js',
+ 'json',
+ 'jsx',
+ 'ts',
+ 'tsx',
+ 'node'
+ ],
+
+
+ // The root directory that Jest should scan for tests and modules within
+ // rootDir: undefined,
+
+ // A list of paths to directories that Jest should use to search for files in
+ roots: [
+ ''
+ ],
+
+ // Allows you to use a custom runner instead of Jest's default test runner
+ // runner: "jest-runner",
+
+ // The paths to modules that run some code to configure or set up the testing environment before each test
+ // setupFiles: [],
+
+ // A list of paths to modules that run some code to configure or set up the testing framework before each test
+ // setupFilesAfterEnv: [],
+
+ // The number of seconds after which a test is considered as slow and reported as such in the results.
+ // slowTestThreshold: 5,
+
+ // A list of paths to snapshot serializer modules Jest should use for snapshot testing
+ // snapshotSerializers: [],
+
+ // The test environment that will be used for testing
+ // testEnvironment: 'jsdom',
+
+ // Options that will be passed to the testEnvironment
+ // testEnvironmentOptions: {},
+
+ // An array of regexp pattern strings that are matched against all test paths, matched tests are skipped
+ testPathIgnorePatterns: [
+ '/node_modules/'
+ ],
+
+ // The regexp pattern or array of patterns that Jest uses to detect test files
+ testRegex: '.(.+)\\.test\\.(js|ts)$',
+
+ // This option allows the use of a custom results processor
+ // testResultsProcessor: undefined,
+
+ // This option allows use of a custom test runner
+ // testRunner: "jest-circus/runner",
+
+ // This option sets the URL for the jsdom environment. It is reflected in properties such as location.href
+ testURL: 'http://localhost:9898/',
+
+ // Setting this value to "fake" allows the use of fake timers for functions such as "setTimeout"
+ // timers: "real",
+
+ // A map from regular expressions to paths to transformers
+ transform: {
+ '^.+\\.js$': 'babel-jest'
+ },
+
+ // An array of regexp pattern strings that are matched against all source file paths, matched files will skip transformation
+ transformIgnorePatterns: [
+ '/node_modules/',
+ '\\.pnp\\.[^\\/]+$'
+ ],
+
+ // An array of regexp pattern strings that are matched against all modules before the module loader will automatically return a mock for them
+ // unmockedModulePathPatterns: undefined,
+
+ // Indicates whether each individual test should be reported during the run
+ verbose: true,
+
+ // An array of regexp patterns that are matched against all source file paths before re-running tests in watch mode
+ // watchPathIgnorePatterns: [],
+
+ // Whether to use watchman for file crawling
+ // watchman: true,
+ testTimeout: 50000
+};
diff --git a/test_tipc/web/test.jpg b/test_tipc/web/test.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..60682be64f34058a3c388b146873d972fa15dd8a
GIT binary patch
literal 287898
zcmbrkV{j(X_B|Zi_7j^E+qP}n6FX0AP3(z{2`9F#iEW$zxm9=W@BRL+s{YVj)qSwn
z+Gn3WD^f{85)lp$4g>@QQCdn&1q1{v@IMy})ZZrvlcQuHAjBZjV!~>k+2_BlD_(nifd<@C^@o9i
z$6nT){<2j{T_7D~;XdFKlNchn8RE@xjrl$8hnh&1yjW2f<)yJ}F4noz#hd58wCQ7i
z)BXFYo9l?6syROof&@Yk=AM8*rZGX~wx2t?z2^CWIu1oVQ_xVIVw){==3W}&AKo2M
zxO+4($J)aZga{;O26`csbiqv=ROnXlnAvaI$P}IgyY2)98<4zZu1n&FS^NxeKu^dq
zFtU>;@33mzn1_0Fpe+fDLCi#nz`_0tf6hAc5n&Oj$>J$)U;|d&RG-%LPH{a-+%vwc
z_w>5~zxrpy0L!dx`tBYaaZDhq0Zs98ifSDebE^0SLbqlD4>di)7*!WF@qe*>6&LlK
z4w9lG)~(mmX(
z)nAd9#*$rWZ1=G9TE5WJ2S)GmMM?uIY1QShb83_FsJ)Gg+;Y4Jm&_H5?P{UFM7(lDF
zph8(OGMc+E_PEXD6t#Dpzdy5;95;bOd)UL;5~Pq|Dm6l!kSB^pW~O#$g3U~SEd~CA
zBZTxzK%GMBCiQ3Emr$d>~0EAzvK
zzB<$^y|Pgwm@0{+QD|HYO6p;i{+MjVF_H0wWinF61)kc9^tAsMfQT{;C-Od;XzIin
zot_aqU`9T@3^hf9*{jrVpE#V-_^%r_Fe~o3^E#ZI9urI4->XDTGD;fk6>~ykFo_D!
zKDV7l4@`4lkE7y{|6?8@iOuLQA0;H^KIx!i>^IL=SHPb)-LlXi71V5BwFMu%?)iD_
zIDBD25xDAri-mJ-Jp(_)MtD~#wV^0OZEY!35=t-Niu_arjv3JZqFYo;j5*(A
z9Zq593v%i4%7{7F$kYe$vz&xIQ52MB!X?Z(rj@$Ye7YOg#XPfuShS+?rt1T3R3zQ$Kf7oMF=bs2WMSV
z`v}`al4*3w8p@!aLOB(@2oWr(pfg%t{L)nHA8$nqNbgk{Ylt9#m1$s(#g|wSjSKJ#b}RZ{mpKcOFS4
zi1j}aJcaH}@|N6J|AI2tCZ2XxSfv}E>pr4+zvN=eE%@Z$B#i^a)A^Fq_3<0O+%_Hj
zU)aJ!tkUdETGMy})YzjM#MCvrP@Y%}9@&hq%Hg)>w@WhjNXBg=42XUH1tj2#D|gx>
z+aK3fNfUF~j*n!MANaoGiL|}PMH==QGOa81KS6CVl@&fk25->l+R}eq2S4D*ft1GF
z-T)QuK9K^#NaO#F*Aa5~C-^3Gy*7~+)pupP<3~temmjme-n!gT4)sx
zajn$D+u-Mw-OkSt{5FAdo$M)t`hRT-5=6Y?esPfo-p^7OQjfXEMUke4sAPI4fa5D^eId^YGGcD|xuE5r%E|2HmZMJK+eCaS#X
zZhX*xZNWTCMhk)b%dQ6BAi!|`5hsZRrRU$fKM;aIp#CeaPBRz?=6~6fLX9Iri|eRXNc4cSbM$QMUCy9dH^7^4eN1Z??0GAA5oxtwu-4G`|)8
zkcI}ivI~-g9){a)j(1#(a~wl*u5pW6LIw=ahSL+sHR25NW_rYmHg3*0P7(6T4(zna
zI)1|>rXXItk4C7fBnjnP%N>?maejZtX7XK#>wRT+9n1SRe-`psBq-%nqAmU~5T4g_
z?=(v*?R+2JjHL2|r@I|Zxa1h&r{HgR
zn1BzqTlQflWWgd+%mxwB_m+ab~2dt3dUpAiY*&!Ht@L#e)s;^6Gbyl=ey
zoua^==h`P>3mPhId_
zQ?nftoIk4$+m(uHt-3qNQECI#BZX2>1!4-4aT}q9A6;03QVN%-grZ9Nh;W^$1Kvwk
zEI`f^9umJrI*l56Uv_%t*mU2k1E#|PG2^vV5eJjcZghKqMrsU&hsDw@U>;WZk{&n%
zcPs9ARpKQ#+)cBZBstA(>RXCapoE2_qJ10z6AW~w7^@H}W+x?N^|~!)+!f9kfljn5
zpilYcq=cezE2KvgK;7h|L>umIy0S((hZ}0uFYH5Vk_|us0Ez%qwWos|Yi(dtsY(st
zY)2@GA|bE=2BbhS2wAfo9Mo@3EQmK5HE@Zz0)D9?x#0CsyBL$?MMC=VBYH;+`gf{&)lM6?iIhEF@|9$lAO#Zg>n@Di>9|?)iRu?;rOnQwi9u#u;*2X&StVGKROymZqFA40+LnTJG6s*qV?=WDnstu#
zlKO3%+%TVXQjxttyWaDMs(MDM_ffvv0`>iLIC^Zwgdn0?(b^hbS;P!aq?Cu}Zvd6M
z+;i@8)Nsn13YP{T;xg%ZBY$aL(ojW|1a|;^
z4YTx&FZjeyc`Io{7MsF^wuEN%VN9Fmj1qM<88(^>$VZ^VJ%s$;W(76c?=*7{d4wk-
z0$#K)Jg(`M$&F}DIS{hcL8(f#}uTY
z7ZkR4jcDD2e(#_8N+#j>!aM~|!6W@>PytKF8*vR;OOrwiVw3e0ocEp?5>qbf(b?4Y
zzue{%CNWQj_R{QM@R?DDUZSLiyLPsfWVxCtbSaaKmXa15O%6A&p!T~{s@r>TBgnOX
ztiFM0D%sP-v8pBkQt2`IPvWFmAfg(m@HTU0pXX;RN!|GNy^cJN9u^)7>{-FwVBAt4
zMiZc@-w|B6jvloov}YuE7D_l4oCQ7eCU)q|v;XLmj%=tPIeA`+!~9lGm%BbHie-~Q
zywq&$z8rqENa6)$7%j$+nfE2OpJ6jvV))9Ivyb}VC^KYUH1z6qw(Oxs#U21Q0teRq
z@*0uquh5(z_Nr~;C}r_s<@tQmb;;(OHYqMqs||1veCN1H_r`_zyZrxv>)wJ;VpDyL
zG8dxcNfH%_f(<*!;Gv;|k_iX_EmKAlFN#z0410o@e3eIcsh|TO`38N%olO9VgXxpB
z!DP~OjffVJVJnl;pd6#J*i#G?iIsRpX;BR1Q;YJwR6baQX*bQv3@8=8QHq%{@VW&r
zx7|m0BA@q!qfXJ_yb)OR4Y$G&SS8<*CW$MzMT9R^YP=f_l;@$KlNS-=>-iz3K)r5F
z;Z~GJi^?u3>XRBOP;@}%rOu~-U&}LlL)@I83XNhR#-`EhEfU27E!;+hTzKN=|iU)s@eqHFMSe%0f)}Z>}
zi8n8Ul23SJ+x_Qy$?d?;4WVomYHM%awVLxGHz=
z$>>d=72mg4w$4KOX;zY#$EP^D!^SR7#0y1o#YDVqK6LdPY{b;+;yLV`?I2d(Rku@T
z51QmHma>-iHpJ+}MNSU39Bn~^6b-Lhip&Kydo~xxAF)ISSfcHH0c|4fC9Eg5+wYu>
zEkO?K(>A$n=G1Rb{Gs3-yo``k5h*zvC2FZZmYawkqh!%YMw
ztxvT+mC9925*NKJei+X;m)5|}x^A>_0n1pXw(#Vu)2{Wj0f>qk!^urRDBv(23aN{V
z+gsuLTncZAhS(RWX#EMLEW$b|#z_NDDZYoTrMDK7GJ{W)Bi5fQqmg`IlT(UMq*+dI
zKE}r1is-2mFykv2UrQ3yGmQ{#gEXa3OCepfr%v|VT}t?*H1UuMKOE;G-AzKf>9SUh%Dx<_#EK#H!L#J3iDrg@;T5{*JU0OW_2NVv41x#@kW2sHfVbp
zG)@+1jN43m|BH5trs$3Y;=|;n5u(ouk-)(3-)|!p-Z%vN>5=tbJ3w
z@EDjTah`rTZeK33@0$P;V-hpDvjtx86ei6hOlxy*LVl_!?#ofM5}S47>@E$Jt`PXb
z;$qLxN>J(awEkW*p!w!th)M$FmJdIf<%X-qsJQWVi(3QdAhp^irTnY!N;1G<}_m;mzbFl~Hie?+s
zvCUhLqmO%?hh78Gq9|p-_ub=u-+&ZJPaSQzIKnEkA*aK1Ue{``lferhe7#!AsV3yB
zp|819D4vPAx;Gl(d4btx!Rg8Dpb$6waEj{F2ta&7>KcF$Jth|}Jetvf=i+z8O0=mb
zRXm?na@iih4xbsAgNa^`&0s5S2aeF-=-zqEKps&KGcXBB){!aVO=i|{#hNv}xi|NhjIliasO*nJOE$xn}HF)?q*}<}c
zdbjD(+)c}>#nL_E9q{i?^3!?mNf26TmZ59lBZs&3_X9W582G>i(P2B4bD;;+cn+o;
zb!4~y4QNE`e6Jb6aPt%5YU>%%jj8u@3diz@iM0Ke(VL3VIsFE>*FAsk>24QK^n-jc
zMRc?XW#X=9Z+zsaPWr+%-Ub}kOI!)goj#_N%g)E~49ovv)v)&%Eo2s@HrE&Pbr(RI
z)eRv(h@pr}kQ>&&?ZAoxSY2jIP0!Y};0~B}?v-mCmA_Oq4r^_B!`%EjWy%m*giN^q
zaWfD20?PqB)GcW5JZZRJLc3%3DqF=?Fq1{@I0f1w3-k`Rd=aFsNM@RGar@vpe+x%v
zJu1*Ce%BDtt6d-~9>v>RWPYyDRp<;T7?TA*DO=MrR*S3u{Q%PPw<5i$p049xx#mtO
zA3|n&sK0)tRYemY7U!v9`?*I<2FqPNS3_|QlVJaKh<~31qtR53aQRBI-DR%AQM?
z)^Kw8L~LSgnmNrqgbhjU$`77rLC?ovUxM;*w!L8O;$X_|40dhAe#YVT5?irT-SEq{
z<8R=XKmGv^M4E25?y9t}a6s%r>LbD=J;9+Ci{(Azyw0lpf^ugQv1Xu5Y4H@_o=gS2
zbPAjhDT9`_g_*S5^znP^h1P1cJU&;ID+~Q4p8?*xf#FRdE@_xa1k-Zu?uTtaP
zgf0WU`}EGF>(%zfouGJ!A6^|3+rry{%x*L$$1C#n;w$tFI`H=eXed2%RjnL)Z>C4U
zM%iH|vymzz4FTmM=xdPx9HSNr0ot!9ml!mwKRZr)jl!Wxw6qe(&6bH#j7MD=l-LT+
zvkl$@mSG~hAuiZ0EwzO+e)p2#W*d0iHJ+(cgm~xt*lZDM<~z+KX{XeVwI-M%^>e3-
zveRv4WkuU&KF9#oiaoG4qM~?0JIymWg!aS%t_|@Vxp!Sm6!$on8P6=zfOj)aNun#`
z(oL~A{$f&56AR@s@VsS-y&?w}F68)lMeu|*A{8xoR*!Pn;Apo6cMACM)T|+Wl%7GgQuS3$J!(eB)^bEe$LZ#Vc&S
zl&L>~P`=t%y1rc!UlT{Dyf`61QNj*J}ko
zV19vZhoAc-9TUgc?~Y?>l0(LN7E_0Rb8Mk%!*ZR?1ZH9c}(u2mv3
z?b|MCsm&8BnFjknd)zF3OxJO@#_P!Ee}l(AAfQz>AEuLuXYj}@l1z=VU+it$2UExq
z2v7cvz#@$~Yl?4V3TU^ku&QQc!O|_82gh&+)bNX~7k~=xvB*&040s42{i;h~*OXSd
z%oUHlxIy}k^j&Ec4mGvsn^%d=4^$K5#1SIR<~5y$>LD#6Bg@WsJ^Tnp;0{4kWgWnN
zyYfP@t`uo|g|zhED>gX9^L-<6q~4Usq;_415dPQ>ON0ui9S`~Oa>x4}^fXd}W?0}z
z=~Ofe@wR|0I$cWqTv)2&c=b}eX>XP4*PXQIB2Y@4z?yzKuWwb8ahrIv=i@i1ghMDT
zd#guXSoh2JkLS_&E#Y4Y1!U+*BvIa(fyeL>5%gBOgLAu6OX{6pgL!*Xmy!`!HYp!y
zAv-X#ceQ)Nj_+XFOOFML6?XjJJxj>@iL&VVKAt1{!5{_flUXRM?KRlhe=xQvRj?>R
zP-BcViy8JW4b-cBBh_tR>@#Q^AFx4l6c%S63L&JoW8A(y{i7w#lZQ9^8lkeOX-{Il7{qeiIr)zZ}Sf}@7^{&XWs+j
z7SSuE{^%>zCluaj@(op&kem;bRl4~%nn*pexKCmZaoE%>9-%NeVz7f{@MyC{JKGy4
z!@_Zj5FSKWgffMSOo*$$m<31T3p>};@?}5sRouU2M4)Y^=ZqVb75;gF(5S`?-#GCk
zAIa4!{SSE=yWGK!&znShi*|_Hr4!A?m1CV9?|6QNpCu567$gZ-!Vs?Z?oJ@+v>!P7
zxSXPYr$3)ndlO(tk>%{9PP1|+ZCb-i79l&c@}HvGc0%&r3-XvKZh84CdkWbWZgmvFC^6D6Sl%KnGlaf*!>B^}s(OJwAMtVq2lEZ)&>J
zzf;Ub&LdidMucMJQie{q=c9aG&?geH($vOggnTa)HXgyC!b?UKcs0-DGl=
z(IQB&ovhp4+aIfo+ZI8F8_|2
zrc34YIU#=D=Ck4sX~W)U)je?7D{4nzx4rm6V821#JDj=8cYCL0A3RYh1z>jyu7@bFAs`^|9#T0^)lHp>}tbm&yy8FP2USh%cX=B}ym
z!MHJd=-F_Ngek-m51+he)Ed6=VUIBECqPBJVV26r2_MpBI>O^(Z^>L9SY}zztp|-3
z*|Zqu0=G1{a_YGIJ1Zb|%nSX(`ZdT^l5Y5I1s0}Ki-rO-Z#u0#n1%f)=8?x$D^oBs
zOwp%f{a@l%u_x10if=
zoPDC290wE9ggg8@5BLu}M!2M=5Tbux0P2(jJz7~#6Vn;e?lx+Xj3*Wxkd8NKt?ZoP
z!)Gl2mlH!{%es&$J!o2K=21GNY71?+1E};ABxH`^x~fOhwxgkb2#
zEN~7pp-;Y=ZSt^^TQPUqlIu7`#UkR5(CRA45}#>`kIPz{*2s
z*<;F-qu`2sW8}=GXoKjN6Vm#!S(0^lpzM_>zQ;BE0>SmtL{$nL-AxIJEWv9Tv28le
zEY@cSwV0&;e0lo9-+w{*dK^bgVF=f7B+}*eTmRDvyvkdb;N`ub86I%i0WFa@WLde+
zil9?>O@>dJB)3r(UW3bwic1hl&a4QW90dV5`UdG<^6a9c>V$Z$gfJfdVBC9O6%=C9
zHwP09GQq5O=W?+^c)h1ibmJL`W?@!I&k+pL#4;2up*D0ySDE$KdVHh><7&Bw+r2l;
z&#Kw|_^X|)4@Kif{qw;;&p&j(pcw}pw=eFGJnsg0wGo)4A>Z|hMB
zG$xgYp<7w}kpo@^yN4zxE2WN1sAW>0@t~yc?IK`Iiwdd^?l(dsCZRL>1+VvWVL>kUy!^7^ePQ3q+v9(H0RP7ewP|PQ9~ieryKBr(R|Vef%x)
zzxh@|Bsl1#JYT89Y<-J?P1G1^2!)Qrw>8{d`>2fIQOnR2(uAvIwRCoVfU{`CjF92@
zbcHf>@Q?68b}&{WARZ%Eqh%&QoRQH%X%@*zHdze9g!?c-U;ZU9>@TcO%Vv{hWZNAd
z@9;amI@Vn+1PoZ5fC{8Sx6&9UkICx6fK(aVJ*_se{zOoi$`+!VY8~mX8V2Y-<(l1Q-KyM(CM@dnEA
zxRl_I9wAmhElV_@qb>0m!K_k1DjTPrdGTZ93~eyK$KngLN=7Kgia(TZ8`i0zOCrSSaTQhwz`+9bj4d_2b8>3_=5
z>3r&=R2@NJj#ue?=A+IDgt`xqBaz8r=kNggS)b6n$Jdhi)OO`W7pawiNO3T(a@IWX
z_Ka8oKC(h8F<_J7kbIi+I%Aw%DShg?w%Ga6R;^7^5=JIA6>fsvm@Xa}+)_lLYxf&0
zR{$454KOyCo_Hs@>E~c0y_;HTZMfSP>&(0^>=2G3QAl;f!s&xYbu?u}nH!RyqvqMH
z*~IonlqL?~{a}x|6L8@+hM!KmbSYTzeZwH-9^M~&)56-si|j818~@F?>3e{t#v$U(
zH2hGl$c%TD&#$`odFYsUbz$UbgDmL(pr5VljfwhGO>1d{yH<(hamQeuIw)!lD$U6}
z)CA_t~{{NE9QmWsk!K^CmQT=}!`mqK1@nOs{8zaKJSBwCVnp
zCsO$yoJF`4uZxKZy5FckB<86zl6HyKv0|OB)4XU?Zq|Ps0N48cEfOHyO)_bWg)F?3
zw?@TRt@(sUME2*K1CpEpz_+~)Edf8UJ@9bc=D-$(FPK`!E(Kwt1#Ye;EdkIfR%
zW)w*5RttoRWm6;FrWJuj+V8DLa^Cb`c4%qH5{Q_=7?J1>^8Waunabe{@vjd7wqe?x
zBU~SZAlLK1z;P%tqciQqvO4Z8yu)JjvxL)IUjr89Nb2WuB}=^J9Eu}WLYFhqMi3!@
ziaFx^H2LcYaI4{2G{m|>0yN5MaImOjdWH1E5#K_-;q@%+@23^g!g2HvYKUr}Cwl_=
zt*}jh$LcT&wO5f~!=_nI?N+-A_S?7WWQp^+~YF&SP5Y
z`Bw&3;+Qiqr09UEcb0XY-jLE!&@~M0?i?&>QR1R-b@l1!{N7+F}B1=k%v@HC$d8yHmL>R&Y
z$zO3b!VTXGoDRYbKjAu(!-hJcp%1+W@qvP4d{d=95{;5S~zM%%%QoCuWtPI^50N24)uP@jq&;n
z$f{(PGMtg0Mr{nE&O(Sxm@xW7S{?;}?C{Lk@3=DOF>m=`h@JTq#;4`vyU~bYn3~9*
zBXHKSt{$%I$~-w$^z}_x>*+ivUIo3&fk}C_QK)f92-KHVheeS+E>+#3eT#Sce&drq
zdYGF;j{YAd4!4wnyeiQ>JRfN2)kwRPW^>dq6YdT5D>|8;sV;1QLAsRj^j5+#O0axtlOmD9
ztmqs#bi$rG7v2StU{unktzuAKh9tEu018WUP8WtuO%T{6`CK)OO>bMvY1uHD7+XM5
z7z~$f4ksD(VsHHMWNwW=>)#SW>fmZ_3T>HwN7xd)14mN!<`1v1Lq`IQV-bOe#cJgL
zTT+3P{$LayU`J=z!H;KIwH{4wZTW)axaN*}1=*91&E<)1yb~NiI8=^7nmH54X#?LX
znY1XACX_LaVd*WVW!};s7;IafLFM1-jN?cA0Qt=mK*S@hJVtP#0KI0TaaLBywihET9rt`|7VL33VngrwCb<{zgG1k+!Ofzs
zD**9Vdw9-o)S+a3r5T5kmS8$d!@;c5fS_LnVb#mbc!=9}@US%PcD2gR&@fIo$r$@c
z76aLwXxLqO4Od1YB=cxwF`h~rc#dShOz%gY+cfH$EzU58ZPIHqf7%$TEa}NZ#Hx*A
zG;a)#*!8yw*l3&54>Q{k@m%K9(%T7xJM4%$VkU~figJf7j1HwSi%%V8ooY|*H30W=
z2gZhQFu8e-IrL3CUqyA$U*L5v>ENf%{;UTCI(r3vm$4ZBEAQA(Pk>I->MoFIUsoP(
zYvV(AL@zwHL%Pkj6n2hF_#mv+`;K4l&ocyNR?irmjAIE>?LBG`lQ;
zf@POnV&j`9S8GU4FVvq0jE2BTTeHGJfgmsLAQ_Vtz3n+T)L7h!?}uFHmF0vvT(>fX
zXV=7A1HZ!!O^%EZDe;Q%@Xe};6qs@zx+2#~jU&{&z*c-a^BhfmJmDi`6i5ncy6^d;
z8>6Mv5B(TsM`9K%oV>TJFa%>2c>G-#KUeLTPIZ0N{dddy>JLDE<*rZ;emPjaDU$snsC{JYje9%(PDqZ3
zfqCFhA(v375JvkeJ)HoSG~23LNP~q|I!AuYcoN7)
zU1c{6A(9f50
zc|V`a+di`V!bNz{#KOF_LiiEq)wV!eT}3xg15BU#C=*eK|7#~M=M7Jk+qKZ)m*O;n
zOn>EfjYg3@qQ+#1WgfyoRpBa7>VTzFq>;XIIE|-5vTdT?@8omt>d
zQ&bUs*cg6RWAjgpkzLHrhqaCb?XZvOM<?
zVKS(as|DIT{D(TM$L_UGHAax_L+{OZe2@u|xF%h6h1zOt2K__VAfrS2QnW4wcO&w3
z3z>tqbdEjk?<{8-CrO!D>|)>PE<9Y5!DOxTQyR~8msk5@X9M6KzS_EsrZG%OYK7ncu|bpqGzz#c$7;*QB1NA|y&%!=Syu519D
zrD_OEsTV-h`6Xmhf6*KeHP5)ym3gS9uog#Q$lVt^BuB$VNskuOoYJnZyb=KXUu5lw
z*8AUq;#K#IyBC0WoI7}UoFBX^_>MbMA-Y`kE=Z+VjHHEW*1#>MXpAu%5ADut!km3$
z^+u!LC|bw#r09er-$ZuVuo@_`Gx$T`U1fXvk{m0hQ<3qgP1)QTsvy*d@Ofp&lK+4)
z>w<6KO%btI{`BP^r_I^fg)uy_-rkkoV{p88^;l`aQSvzzaeSxe{+W4_*AOt_ZL%6`PLB{(p;Yl2Pgb
zJj%)$0s4=lC
zIoYFco(=KgyoN%`i%tb3DHh25qz4NjnbFC#y@{Y$klSOiH_f_`CoUmDpTWh~4zEQnr&K-!nF0ZDoj}
zAvs4PI#$_i5m3z@1&M^4KG$WY|40ky7z^@Bc~HSJM~bw|Ri3#J68`pdb21rU&utJc
zPk-yT{m`a)^9al{F+WTlq6D?cJ}_ZtA?k@(Ai1{rs6qUN%O+6W2<`
z@G`Kt)Hty=VPxXr|5^V_ng=04~&OLKN6LWwGZ@YWIxi-{Qn_HHSEQDIixz?^r?O>%{b|Dx?;&)!!et3HNX{gPL_iXsCi-v@upI#YrTXRH&@>rX3J!bER6V!s{BuOJ(+9vV)?
zq|_yIhvm89^Lc@#!55!-Wz<}1sfpGPMPHSksGSm}fJ#zZr@a~Kh=nDhx_3PjI2&bL
z%~R^u9y3HoLTz3@LuOh1&_(z~TbzTWmb!DldiT83259skOr$0uBbtc4(`RpWCpk!J
z3x?ID^{!gNC9Roj*`u|T!74PZ;aRNek(o|HYP^qJIkX7%Z-Z~xP)LT+SkUU>Q^78v
zY-ao(p^H{R&oU%1RyH7p$Sn8WS6H?RB@)4TZnrO3N6A?k!kRtJn~H0aH2NEc4UDRq
zR9P(}Dpe+PLaiP>_fsOmbfl4L1&E}Cd8B_^Kjk=0vwBlzbTL}Sv(45XC)l4>#>mWQ
zuA!HNQJm3V%@uXYf&-RMg2D9NJ;iyfxQeQSSe$MQ3m68qNGKF6V&p43VKvc{7tqgU
zY(wl1x+&tZncl}XA(=IF4Q(Xj1c}%F6n*14T2cjaGz=s|t@N{i&HfGc-X~h=CF%R^
zcFd45g;o>4t~WvB@{R$xl08V<+PNfo5|=8+t^V4rH_WbY03(76;hk05EGCOeuAjY2
zEZDiIG!u8=7$xx$BV3pntw>!uUtmw+UzLuLPi$b+oiB;s*-cmvM|D%oMySwgj(6r(
z^R_oQzcA=lY4x&2%3$hYD4GAd+|j!RQR;Y98o3^(3%;A$Az$vUT~xMqyJHtwg=(z#
zMsde_Fn*!yaou3C>FTDeRR6m9-01FWZl135EQ|)|f5Ix13JyI@b%$tGP;RW%HJHET
z*$&}-vzn~?QESTA7I#({Yn0~tJl0N*ZRCIk&B}FlD2KR(av^{iqwBoj$Uj$crfa&|S|0jKUT#Jcld+Ct9)8nW%
zu|uH!j3HasAF{ELFc2XFGXwv}3eM&rwnwc=M>jC!28Cs{dw^xAh(ISrYpWx~zj+$G
z6blJDIqP|Us(YZ?3M+Xw^y4=aER8=ZT5{d1Ft;r9OCWSTU8#RWi$Yh3LFw@1liz2a
zO-zKX-Q(3>h;DaZ^(0e1+9|f2ml?a9iYt)`f60U=kluzQa!#~pH63lrR3(yjB?^vh
zBx#`GDQmAUF$r2VN5yKsfB-3hoS~*2i$g*VVupL)YwB=8hc{&r!us3z;XEeLV-IHN
z(2kxEHz?O&plxs&)dS6KUVIaag}t;$xc;$CnJ3B@V9ux(5pO4($w`C=OkzSWyhSY+
z8{QNzU?VMTzFVwXH;{)nTPVtmVvN(*7ZWn|YLZjo^`qmsdMaH|#uKRiL&7D=z@w8&
z=_X?hw&95-dy$M&&S7%9)R+eEdhFRasAR33A*&ZE=T3qSY$6iLgdt=>#)Wq@7S+p3
zB>_J`b6X_v_ubLg)BZ%oHi$VFK#z_1CHu;VS-hgMiR&R!eimle)H-8Qs@1>v*4_K*
zgzT=|pQ`q*a^uxYs3?`ypx5QUQWLYB;j%>ylUY(%d`bHac)7QED6eK3KI+-mmMcyB
z@>-}CHb-ryQqq~I-KRbEVJI~pU@|8i%zsi3xFU(XR@jX!c>%?5w?G90l
zLaqZi5`Uj|=$0IpK43KC7gH~ZaD%p9m#AXTs-rSX93b>ku|wjhZ*$F~DpI5fls}Ue
zMpx|!FT^8NFjW%|9jWTGuoiuE~i1LdJ
zbegI%qjG(G^;#WpGV1jelR{>=T}n$FaHXJ^Z}g!y$IQDhHEt4>UpE7r>5>te}t?
zZ~BQ_y{`?HON)^9gR|`i>$>4H5n~gR2<%p8Zd4%v;prx;R-otglki;CUrp92JvKs_
z%DR6H=V#$Vp_2k|atG=P8BdHZu`qbZpMSfKGNYY*%7Xz{VMyz|+Jr;lW@vQT$UYvn
z@e1ouU4se{YLU{xAt3n4@e7TVE}yJIOo7-wy4pOuvWJ=9Ui0^lyXfdLv7x(
zQr_Dtwoh8DMmYZcyXiXqhHyrz+`gWPF(ta)k%zPW9<72IUh1g!Y-B1YN{MS`xEHD`
zPEzYJgs2(M4p?I6MB*Usp{}ge2}k%?TVeo((oc$Ha+U1}Au6k#)EC(@gK{J4b=kI0
z@zA#e_w~_Q(`-!?2@+I$`XMUkb`PSv3yPRKFz5%yDazWLX0;U-t5*>la>6sQl5;^%
zuLHPEgBP0cr%~7G@KwKjeU8I%wA>gz2cp}$&amd+yn4d24EIv7IZEbqE*b(Su4)?!
z)_~VKPlNpL?6o0L>|yGQC>9|N1YSP5HTPKGDz>9sN#0+`<$9My6x>?#<&r_U85A)(
z=~bXzRj#B`HS|-~*-IX$z}@U3R#X<7oaKkbf-8|yfB
z#%|VRR@Pn(#-}{S3p)G0*G3<(j<@o)2iNm$?1KtKTGw7nAr(jwa?ohH&419=zNRni
zXJ|J}6Tb*fJz6-UQDy!L8IBT)3yD3%%#N4=2na2)*mc%cX#d@GM8*{slEK?UZsF&(
z$Nb0YL7(f-`3fCq>zntRR@D`Cu$W+b=a8Roy$c%C^%i%nnyBmP6qyB|sRt7y$qi=_
zeBJXIl4ID`sT%CVB(V1Z^l%WFnX#5wfgs8xMvXk%3NVV3VRO?n&TU&GqJTxH7nLb&
zf>cAQx`9ORo2th@QqUoGNl)3IBpSezm|ZH)-(R@m3=_+%d~Ij=O`4u6W#X^rQvMXy
zfxi_fJR2^BkancuWc?|Bf6iBWFxh0`e6kDa`%N!sgUHy@6y4U=gPDg{q=WU3+m8{Y
zHliWMAu%hpg{2Ddws<7=>MW;-WR7AhR+>WG0Obo4uGO%DAd*R?2%PGwkaSmKVE2h^
z_(+M?Si4pUpJ+vhdvGdBDLq|Gwxc4rqlCW`l$Bz>50>6Mi;c=;im;{zAG4|hk$ZDKA43C;vhoD@0T-aCg3zrUPM#3Hf`pV+1nb4vSw&lo
zRnVM|p`wy(K)J6iJe30NN^3PpCO-<#THEBiaT;B&BVc`~>ksg3LGPyne8}4y
z=XFcopGsiiKQ91p#4Bf6G)q%WVl$f?xe^xl*ciG}g+CL+ObcZDZ=y=GgPHC7b?Z*Q
zm*H&PB^Ub|ILcem@gq{SWKz^F*YTgpj@t{7s#d7nHJI%W2%oiCU%Ud{FSspKd9?m-hwpDJb0(O*t!U^e<()%V@;KWPz(JWN@;s`1VCDbHP#ys419#>-H
z4d;%VGn+_~aLfh7
zBpxNl3gXD;S_k2<5+5t@3LDsuerT}&aH`9+lZQAOA?7;}LE#Xl#12e(ED9=<8owfq
zLr~nV;5m8&54t@w(Eys}f|N~^#wso2pDv$|q-zh3c%$ZxyAjdO1|1q~)R)#xwM3Dz
zUaT+F=@`{CM=aNN0+%NaO5jK5S*lm7Tj781A_*$!o-6)xvi%TU-+T`3Kb440Bpd#J
z0GmK$zdNh{1po2r{}K~3ezb=)ri2=-SPDuf(HF2cxQTXe2qA_rBZCv&^Ei{Na^W;h
zxQOH;iGvYfK)_3cZ;b#-SHP9j1tEoOMR+j0kH7ii-{QTy{{~fS2T7D6mS=IIbs4P%
zjq6*VVm_Vc5X`3=AK*s*G4kF3<4F#!TX^fzJNU1D@?XQo6a3lL{}Wd?-b0iMs_(#0
zF8&dI^6D=zhgB}n;HY!d;o1Mh4I$$(+LW7C?ZZwEfEhrw6F)HN0@uy^__L4yAC$U}
zd^E3-A
z;to3LX%1mi+dA@%Q}7jyhALyJFu=A@6)m`OmMdr{y9bzW&Ed))yp2-}moePYNYFt{
zK6&*4tGQ8{@nB4C+N2wyVYAHub=1F>`2)M?QuwrmvMN!NdMgc^)*3c&lh8Wld^#d~
zYMh`$hITr~{e!RY!Tn!yhmN7zXEl!$!>9??5(&v+IEYJImWafMFhMQ5#`tXeGkmh~
z9t?K9L2xse3i!UMzSl5V2_of}jKFbEA#Utz2$|GSQG}5^z}nB@B3)3fI(X=5P<0)-%7M~OQ-t2ncA0T<@3K&li@OOW3+r6i=TX`4tD;3d#W
zNX1#75S{)h&a7NvH87%cRrc8&n3UKm@1i1G8ZN*}
zxuUm<_fabgkx=}30yI^{)5208g>G2jVQ)0i7@+Uao`aViiqO2;kTV0IH5%$V0bmmgX!Xm9=L{4EON7z-n_}3r)
zE39J&NoG-Q4bVj!fB429;f*stgeK*&7+_|pgF(HE$ylK!PV=!x(=&J4Hc9!9|$1@t1XYH5JcD=Uh=f&#
zM&2y3E76>XiYJeh&Op`Tq|VcN(>zJ04g5tk-T`?F+G(xYDFLf^QOCPj5_z51(TwpBiEg
z2nhJcjc+dmOgCGtIb2A$rS<+cZf)PfPQH$G)_PN1YjAJv2LA88b?AcHU@44>KDOOM
zjM4#n^JO-}`Nhll`D=fS#hE3nk8a@qXYak5~GqqwSC)%wQ1X#o^3HzU6LgzK!SuuY8^UUtF?2kX96UPEcVO7nwa%cXT?EZB
zRGrYpjJb11W3q^NjGgW_iZL+XSj5}ce~e3=mr%D_FvzllMvI}(Mzo5q@Is%k5wEM#m=Ts~4#DyAscsWR_ak;+R;Q#c}w#0)K^fm`+)xHkViq%9LhZn0%bmIqK2
z8WWTGuEFNVYS|c^5x&Y|1PZc|ppXVS{BasHJAg*$tP0OEMie>mpa7#%lzd_R#3iuqiVdfKX<
z!6|&rXw2a%n6s#BDx*3gAugN(Bd;(IjYffS*gs7m=0-0sit=H~#)IjioM_82hNNa!
z6%$!JMiP%$%}nPkLBtABQ`jG?9OYP6?JFV9N3p(xH6H#l9)3Q?|a;<*K({Hu^ZPYXYJni(SH
z_$L})=RC(I6Y7LhM>I+fNptM=wsCjk6Fk~`f>GhYt=EuO0^IlB$CIs(q51{Vp2v$I
zdGYm7q&^(%9${g1iS0SG4A99MqSkzN{L&##4)+lS0%4%SLJghvDlV+P
zf;H<3gEL8K&{C`8*?aoBjeJ5Vlg4Aol&vyOBl>9Z_8^Ke1cu1t09k$rJGzV+ZwYm@
z<&&H$RP_=qqQy)VC`d7_{)L=@vcV!1tYcYU6l6R<)1ud?=7?O6nQ4USnYIBX7f8KS
zB9`T^v~1)_T%pm(nPU3rTFA-=6atQ@VXbwU(?`bnisX!bO_6v=NgZfO6~ZENe;<>$
z2QL~U3L*wiI<;j?)#qRlT9l5D*0rZZ2!WebyeD$P&3TU=jrE<
zd^442*9YAX*ph~9h}4Wn=vBjEu@#GBl+pk?PqDFmA5lDjN}`T|imKYc^7I-jm`&m)
z7Jy6fCMkV>>M$wg1U~TRaxYGiOi?P}3E4GBId7ibCiT-OUw)q?)H3*(Q^QLCB-9jM
zevHqy|A6i81Ej>MbgPIG%H(rdU?WOfAr{V17BY#TO0=6y%1L{bFdwwwW|~aGH@Xin
z(Z*1#I)o*;flqq(@pqs9?<}sN4UNX*a;adZNJjD~^eO+X$YGa6OR@n9W7;TbsMag|
zHwj;eds?>|`047;P_^rvctXeHRAGEx=XGA+&3bVhz@lJujgU6F#n#*^Do%|pM&d~T
z%_^9`V_6zZrwYBH^4FG9V+CL^Nnj)vF3w)Yjj7vc=u5mfaO6_ClNZlzUU`f`+#2na
zbM!4r$jB!`pV*h{6f~fvm{gxaqEsoH*Cep0Eg_G#bcmzjAx_3Uz!fm-HiTN>$lt=y
z@59M0O!^b3sSe$+K-$Mj!+}MfY19Cw`}pMWeHe?x!7|Yw_P8mtQ-_t2X&U12@Bnuo
zeS(310NqaE()%(391pf&)iorc!i0ZXElJVa&>X@ikrX~MB|sQW7<9072WsIUO;j|h
zO&F@fLW8rWtcZ~%5ppt$jWZ+@A93JAROwh6h@uJh_6{&6)JhLg2u~XIDpEO12ve4S
z^U*$Q#|dTB4fet#0YMsNya)F!K8d
zqZkHRz$~rf=IUE$+bvWiHS!Gk`??}dU|2*1=W${U#rG``!lSqng@wtRnIxU?Hi=EZ
zIX7G8AuEiGlD$n<9>s*yI(WlFWHj=oBa|miWS(##*LV3L1Y4GnXi12K!z_q4U?~R2
z$LTcY7Lu>`<@1H-8o|@A4b$SU=^Ap48VgEt;_Bt;nSnJo!t~PUIq}7x$8!xKjZ_Qr
z@FV*YLONN@BN(MrNNGGd!IRDV2!lS-EQB8ysMSrhYaNbz%u8;fD7~hUOP(R2a9_U*
zDi4Z~QG3cM5=I?lp>m1~P7m`b{7DvvG$y8Lftp|t9OC0g@8grr-ywH$IL#VPMiUgd
z$7)`a4eB*y5i3PtDD*p(ru6X`!f7MQE0_!erfF1aE>t~1rLD>90}1-QG5SXv*tC8Q
zoy03VEfEMKioTQ-1JyKPlYc1LtJ9`g*ZI8XCF-v1tBB(W5m`&+0aiO#@b3H%QMDRU
z;r-{D_Mg)Dk_?VDQgB(<`VpLAX6vw29G4eY=DF9W8^B`
z2pwdkE*wUft*_$Mt8Zg&;VLU7nOj8TLRFfza(|}CMW2W&en5G-6T4*3LGfIARif5KWlFzd$f=N~)SqvpJP%~%H
zsjNcRb10FEN@E&+Yk~*k5An-~zrfbXQ@~9*la}xuYT%Sq`VHnv_KnAz7JX{$M&SUa
zVRMvL8f2VwLd~D9Yi!a?1U6ojM>=jUy@K0UUx%aH=u~Df1`*sX)bs`_MipwV@OVGt
z23RUrDe~`>$Hw6$iiV{Y@j@4&CfYY&8wA{t|2PTDF~)xHAus%#uu-Hi^%_=YmvLk5
zWi&)X!mrOw7{9g@N&L@Gr@Iy63+Sjqd@dpFbTSr1
z3hY_D!_QW(b37QopYRO`d{40-z(+!)D{*9!v%@gURsF{G52G}~@
zz&PxKwMuFoYF=ey_4ZsFmP43n!pdE0b`lne5LQRge)>1{`KRG&EV@xjaS3@eDv1eG
zsWMegoH~yS$z71mt%9^jkdXIgkRtYTnbmB>$ovGhJ!Ay3NCHHmiYQQ7SVI1YnrpG>
zgB0H}bR?Pvt6qm~Qq-X|osPqRAD;znJZ^TN6awKS#5kD%N(Nn5_IW3f4dC<|b^6(nT4K?+4P#Xo-Z57-<&g_lh@*RpEQ;pMB}!_xd!Q2J+*@^$dAA{cmY(C;sdN=Uq3))@QF`FzF)J
zLe%OmVk5$WMo4D|?WqC`Q;Hq3{pTHTQUN4SbK)ypG}HwWXH*F=SkpXmZKzVf4ATVMFr=dMkfNUM5La?Er#
zg+4Qa&Q5uWr(&9{Y+iD1e%Sg*=&+p%3{rTiDgXaa_?Tov=wz6!wj|<4&sCEBDfgJs
zZJOkp^7Sa+)G?jDR*t2SFHX^NHpJcw?PKBjp+w{<`H`yrGLoi9P*-^L}V_V!_X?s+2gD2x%YJT
z++`z6O&wjcs*)p3LKI1Xs?ou0YZdJ(8MR9erJOY6P>k}J@yK}^m*-!DV^&b;DIC{j
zCn{3jvw^(oQkXHaLPJ5Zk2;A!B8#^V8c@e@pvF0$M}5(J0_DbjGl&0e$8M!PYM
zYNG*LUxmSzM-xDcp^-A)ESS$3CO(45XLUVsT${BS<^sm#W)Nm9kg`lmhH>Tzay`d%
zbB@tO9#JcjY;ekne7+Bqi2OM^cwXoAU8onQ0_M?^+!51=&@fCaHZGuLPIF|FVL80R
zrkmsb;4@UTCJ^ml*bCw46TRqq-o&zc8Icll^IO-Nu=ENdj+cW}
zJf;QnsdoDt*c)ymDFPO)5L#FWfjCe8GUAa@zHowJwvWw&dkDfIQy%HPlMT)BXaje4
zegl!%2qCllzW9(|e0-UOL&*#ol)PXQX=*$|C16!Afn!Rhp;Mj3oc$6U)s^XYLPjym
z=I``@WP?jhV-7i9&1r0@C>Z%YY#iOik+O^4sE5N~AIpt3yfXie-n$6bLTITzKx_NH&d7um92=PNUFKn2%0#6&m{pqKl=T0nRZf_(=WpQiollTZst2jYNqA!_7$zg!+xZ0dlTQ(OWPnVj(h7puW8_fN9VD5}
zytuReuXs+NQ#OOt%uzxLiI=Deqe#5@<{EAwf#HxPPD_s8Y15Jg(8vQ+$W36&TELr^
z{|c|9w^?yZ`O*qKd1!D7h+K#A%QsS8`w
z(C#c@p|gmE&Kf%P6;w>(Gj?Do9TtQUs+wvO^n*R@9X(}xtTfc%q&ntXt9b3okD&_{
znV0kESrrX3*T02&q4w|GP%)xLglL@`4x^|t=SMF`
zT^lDm`1=q4I}XQ(7<*%=xr(LwD&D;CJ*-wQvG-SyC#V>f!Z@%j}=*W52hBVRX!i18W@{
z^UVwVIQFW8O2y?2<14Kjj8a|H4{$Pdgh&bCXG3frJmh4Ka5R9(b=0F(tj}G?h1o@7
zi*f81>8i;>BGCdIjP`N&$!CbuBa*J5VN7ApT*CD90$hds=M?tlqu~8Uy^fYS#R!s8
zXk|Dvq$Q7Oqfv$!oUY%gAHxZ2s25e#l^R4$Nd`?UR4?M%+$&hBtnh1@WZkcX)qm*-
zGlE488a)?MjeX=uwM-4sZuc?nZvFu~-eYcZ^E%iIF)QR?ucCEOw~f-OIAQdnNE9A{
zlEE#ZClsRZr(@jPyMqrle+@4mLeEvWq5|ItFwjO&i?3h!$f(+mZMbBcqWkDh_82KK
zbel7G!_>#;+aKY{#$&`$g4CFl*T(Zu(^uX$qsm&ULJb9{q^M4TNcWJ_t8kb|vh{dygu~Gm5)xdfC25X<
zB~6Y^i!U9QuWJCAw<)-ZCx3AI@lk$)t;<~=VSgLl(Fs&cFy*w+XfC7fE~Dix
zVXnQ3#m*{P_7Yswo_o*~9%y1P!pPghUiSeucJJfxcppLNF?W&7uY1*XEKJX1p*;r$
zf9AE78A*f&e!0Gk0G`)*o!57#UYrPo77fLOV$qFL-gpX%3*)(?WN-VShfklrk3Vhy
z3M%n>>WY
z)n`1wvqejlixZ10IU%IrX4XY@Y_ztvTCO03f4U)kww
zG=*c-aL_$KchUva7$b}j=H&O6$%7y3JWeCyU5@t8fIKRoSgQ2R%VdMW*{q{Y+&lgV
zfii-n=&)iPv7RBYvTq!9k~-ZY1UIb$n&eh$T7^}vNtPo^Vhkr+$l{Dq6=6$sxOi^X
z_&jP{UR;RH07BVu?izDXu1#!s8aIIRAXk?sl9=M`u+$bAGC>z%2
z&P!fAqjc~qATFGq!6u`3`aUm*sS8*ZMe13w=$I9fS8A#%i&B)*3o$QkfA}6oS0jB^
zM+w5p7`sPNsy>JWDwK+zz
z`K0D4o5>eiQ7ZNA=kVQU>h+?e8vMC)bYACmUjKUQg*AXFle2|<0qpgb^C(UjTIm~?
z;>p%sJlwg1yzO(tLyE~Ooi$usUV=*ykS}=T$&s!UMMeu=Aytb!wq*iz!Qu6J@f`cV
ztm3)fU#w?VFsF+Z{bnk8*;^8)hSiZAWKU7JWN(q6$bH#xC=O?Aaz*m7!ra5Y_5?e4
zfWYV>u^I@C6T$_M#tLWC5&}kCuQ;NVi=0y$sB0ZWV>~(9z(#)?gS-bcW3;9_c;~`p
zTRRbjT08u`|I7*>tZ5Ry;rfth6F8Q`snRxb%r4xrd
z3mQ^Qhc6P0$zxDW5tf;;r_h9{$)+@9BvF@XA_fo>mn#uq4GJ0W>IFhE#zg5Nullek
zcR0~tMGmaQ;zp$q&*%fE=bsQo8k^-5S7`dI*syh%|0WeB85avBMnWDsnaP%=ub3E5AQ-YbcR$j;gkO*oY`sAOlt%}T-^b7XK#
zqM7dK7ze#QoD7dJ==X7i4IbrEY#F@{O=!rx7-32(?G&OCq2j2}DA^^8kcyN)AA-b7
zP908@qQ`9HT8eRegeW~kZhp}o=Xo2L)6EtM81zUY-IOi@BZZ}FFq{I3pD>Tgk(K}e
zAOJ~3K~y@IB#A`n)p9KVHFb)}rR;7~Kux9EHYp}}wD}pH?%hXjrfiu+qJ}hy*~-SQ
z+7Oz_ZUVJBCF!Ubj{})0P3;;zR{6qD`@f>(B{FTK5Nz;G^njz_E_)anL?xupHZ?)H
zZ2EZ05Xoko%;rgy#Y$&taOvZoh8%%NpPNY%;&9lthtCVZaR6kJqb
znN2jROK7(jv0A%{P}=k?vG7dHYf+P5E-qe4@
zit#E$Dia0bEMY>tL3iqSg1V!?B3BGQf>kkKQm{K^(5mK{Z3Ecj%YjP~X%rQvtDIaw
zCaSvaaNly8zi&|5Y2l!jwII?QlVJj%ktWIocKI~Tqia%s9t6j@bMQNSzVpx6>)*jR+Jj@csMS``
zsa%9RqvI&vKrjLHtb!tyD#sx8dCVCZ0mxBg#=ePw>~nC4oyjlp^M?l*h@aq%)t_Nj
zpGN4V&05wSYMG5s$}k7AvSa(%_Jd
zmT42Akw;?&jU=@^5T_nALPAPMO@R#)*QQ=ZGD?wY6v5_l1e-5HPTa7I3e21~7GUDP
zZmBS;hF9*-c1112oC
z+L^7-V{Q5(%*;j5P53!bpD<#TI4Pn=3WX9JOqClF5h$o@EiC8@JSIkIU>=ARzu$fz
z??3n_9DCbP42n@q*hf$$PeEGMO#_Lh;PB)KW|pB&xw0O$c}-Nz3z$`JGGrYD+sMg8
z*dbvg2Nuqe(%R|?B+bq0F3cAj8o?v4W(R0+xo*@OT5Ki<;y;rX&=J*8~(M0;pyk1*+`-
zSvRIJU7vxi8l2ml@jXXDlIjv=R#Mq~5p1o}MpWabFiJ)kjrQ>I?jP~UyNiZZVXh}l
z5@ZCX&mNme(1q8%kE5dxp=Oj)lyO2!kcG%Z%$e6jYoU8x7SK+wKw|sok57;V5h_{*
zEoT~2_B`gB7jbFv7Urf_7+rQ!3yxJ|Uq(U_$y}RM$A+d#>&{exlBqH&7Y=a
zkguHf^E$8d`qy7CP6R^g*&>lDI`a9VU~<*q#sARX!M(kYu`&D%Ug5#QJO|%@fBiq;
z)w%cB?t@=B`&&olJA13cKjgT6K38@EZZUHJ(H!9I#SX7_s<
z4^5H)JHt)n5iNoiqCAFI46%Fi7+FS0g2rTNS}bVfOt<`$IDcotVi_fs6e;`h^@h8!
zYz-ORxN{9h(Kd=O<4O!SE0dyB{F)h^uBY%0>+yrUf;fr+>x-((FPBWR5`%bA#c4Ki*I46a)luvvdkd#%h!GO543U@R_;KA
z5-rl49C;-)hspd#yA88n(W6k2m;;Xo|_(Rm~S(wzYBoplWk0Fl!0rv-=
z^XoBeow;fvPOvt!hAXYxPzF`0$h1dDMFO8JM92_W(-?&+(i}ycz*00grUg?qWq!7n
zBUe&%qa$n%9$=L1qUuyQFPRkO7R8b!bR1vR5>hyn_F-Es#Ic2*H^w9#FhZ(JsXw2W
zGjKi2SvW$zoNW4@=OzD6lt~oQe485*5@Tc@9&dk+z3vurE#dPYr#UKm4T@2Nrxfs#
z33~z(iaS$Xz>;-Y(g=tWP$|wND-^S)3(40-TKezkK!85>4nK$Q4PjI*cwq`9Ffeb=
zVtMW|rfc(T@01Warj%@j*nmo#7D2WvF^{@cHWXKRLs&5@=rmiX8Wyav4m~y4vWifC
zN|G^kjgJq1+9ERWnIOd&UgYtJKjynk#`(&;`AX%ni+#ad<91~M?RFiD^+l|;E@Hm6
z3KKQBMhA1XRak^VpVA&PiEuNAnW7}g_9tV^oKbU0EPYnzXy19A*LS;KTmvXF6m)w_
z6d0P#i#~CO4hD~LXZv?}y#Eo#`3RcXM8#Og+~N&X=aw)uBIJROsF0!lQ5x|mrK(pk
z?RGE;RZNnCQN6;UvFiv=26r*K-{Uk7LKH|x{Tx+7
z3zGnoAm@5G7M1c6Nf3r
z&fYG@-M_`y>*0+n-^YwQi;B{LCEPNjUzI5-xfIe+gcMKbQkmEgX+Fl{7?IXRV)WRv
zF;fPtsNFu^g?Bu{!1Ewx>XJ+KHAU+hrFl4!$}KdbpNd3+C^MEFDJUtSOemmsH1d25
z$0z7Cm!Vo#iNg}#Qt;1Nr$`1XnWdtZ#-XhW`3BCP;f^_^I{M2+$YDP@!f|wfy}=HS
z{9|N!2uEb_CK1Pvtu0-^?Q3u1TH`f-yu3_aq0H&nts?8ya5#F1G#?^Ud<=Xa8{IqT
zRHm^sa}%ygi67Hv&;CWch0}^_l
zG6PvAWoYFFM&V=~LsrKJOd4yxfgZe6&J*_m0i8!fnwiy)5hVBsaq$H@&1GViqCd_hm-h-M_oc`aY}le`q&*j#a{0zPWp%N;~@wCN8=EBs^Rj&
zOL*_re?gk7NN9Ubi3z%bLeJ4pk1+1_P$V)JHa+p7QR;@SAva7;v{20gqi~3a2PZIu
z58pq8qIS@1&q5=#&+rkEtRjsh->@L@iGZf%5Ru%O!yqK3Kq>-sqdh$Ce*#bKaUwus
z<}gPN>is5^X2G6$v55Kmf`bt<%D7bvBu0q7cL?=!7mxQoMNjNPBlNKzwBK14S-fYm5y#YEL=V#-~Do>m#gkPWAc=emq!hzB(IS`ZScVz8
z-2rCZFedb;t)2v0cQmA;P})sDM>wa!gUYYRZ`N)lc!t&!M&F(YMb
z0!5H&G>5CYxHx?kmls}zi#D+XzoOdx>NzBljmlTePVYYa$qBl{him~uO`R!B82U5E
z!$K)rK_T5Usc}42(KBpYNM}v@ydL-m_~WDBVB3F$)F#9-L+S}!oVkV@%Rj)xCobe6
zMq>fJqMHG|^i;}OBV`K0Lzq<1aeE9HvoCNz&njyBv-^0JV^1s0~9n{S_^pdth
zyDpoQ^FsO%#)St{)c=B7|GduYyuM5I0&QS2UUZ5=CU#Jq*51h$HV!{UKi+^*HQ-dI
z5%>b#lRGG)T^t;INbMW^Y=l%z5EAh%6u2~d5%uXS9L%3a2@4ODR7d6uWFc4#Ld1JQ
zkO*naVRCSYu4=Fi6DDP4+JKi(>{)`03{^D+lVFT=putRl;c&>K;%Jm1B7(lBz-XD!
zNr7UNY5b~W1k%zLrxC{f1fCc%LP??YwMaq7Fp+B+awdr@Jg%Jt6E;1L@(`I7^SCG5
z{sXi|fK&`I83d>+7x30gKgApC{{d~Wz^dcFdhN&Xe1ZCG1&@zD#>bC;3n$j01|C-G
zmvDROM_6z!BORzbLagg`#99;A=D&wL$?@pm*O-hq5gQ{$Cj27$k8vrx&CND7ixlBjbE<4R!x3_&7~-2tNZ3yi&y2gv
z#`83JtJDTSD+Fe%vsmd|M#XHhNQ1diG~!Y+MwF5?V}&M-*;rppmqn6t-mR|7X4atO
z1d2OTWuG-353x5QuHOcB#!s*}c!GYgk1!uXGcxFrj=I{y^$Ry~bNw}}wywb?KgC2K
z4Rf|ABCa8;qyvFBFMl7po#B6elHuveT{izsM2f@7Hv9)ZPA0p!GW`aYr`9lApFvG+
zumSVgu_}nB79vOkzD{*RL(cx0ChB$v3(Yl5)jL>RSU{(>fTg)>_~7wB;r9>rnT?iZ
z5wy&L${E&m=|vKYQv*UC{l;0!2@}Nsq8?aH1gG$H?0K}Y$loZ^Z_yNpQ*j49OC3#h
z25oJT8)J&&bB#7jA`5+C4T@(4%A+sN|>X%S)L_#rI(28**RxKOt5
zI1wQdA)XxF#oevDwBAV;hv+)F;7gXj?ZCr`0;^bk8oo9OrYIEf!%knE#K$VH%n
zW_=d3?Pbi=SJ1A^VY)ttc4Hn>)-=tvfxxFYAKowskUtV7i#V_y8=*bH=LdJ-9sVD9
z^`%#Ed+j!A!sgwIKZ)1}P`4y!QTb9d-+7(ad42cm1vY>=EBDxOp2QNm3dK-(?3Htr
z6?t$aeB-WFVg%tHCf*KGB*@K-?&0Y72f$dxw9{cPyVUcch6!3m4RiJ!6lwx83X)g&
z#xD%Ps36P^(#%Jg9w3ee$oxa(nFb}IXr>NYl@<~sf^P@VOarE&qMA=QaGeyt3(h(!
zn1WWg$O0j~1}z3E$$1N^sG?;p;$r&-GR;R}6-+=U&pnDq^M!{X$eFa?bZ1emHeh51
zXLA-t2FrG#*Ie|XrwIBUrwZI&e;Yr&`k%2>SzuFK#hAg|jUOS8JA4yYV}bj@J>9Y;G<0)cQ<~Ha&)vSE^;
ztBgjeWkgg_T6d_bJqrvDMqeBN1)}8;xok
z^-7CPwN1@ouAWvXtnAI1%UH+S%nj7+26P?V1lm>|O=A}K-479{W8`FxoGQ#!%XoAy
zqfXMi)Iw;Bpw}x>(
zX3Lzq*~W~s2wjt(g_JxG3;EgM^UG*$`8Z|zNL<9&9z4cpTOVU@vI)0NO<@5?sbZ;d
z5m#ql#+21W({gZg@iw+MH?Tc?0JSP0oE*F3C-~K;{|HioPRucqCN3}E#I2Rvm@^kSZ&_q2Uw4X-#@D=|JR8}|9SR#=
z6XAn4PgE1`*-(?kjQh2CU=5e|DSA71b>Wi3d
zu41No32kc{rtW^h#V;&~ck2>$5Mc@!m1SIc=|>1`jkAcoNFX5VoZ$B;9(R!?ecYUV
z1+%qf7!Ktn>&R&dp!fDo5dC?b*Li&>>jgG|bmP(4N`(f1q)6*DJD6(Bqc_>Zco4C|
zm#jJ})ix($5V2j+HE2}}dd0%D9U&j8n5$lcgH=Z1nobpOT)d5?nOP~)NOD-@>aT0a
zh`_IqxyCZa!3mxo{u)~+e?XKBp%porwK-gveI2VaH=t%Vl41;K38NHg60rwZ5@c}o
z2BvFEtXj@x{C?RSsXSV=)hgDfUP1MG6}GGLPFoZv_PrhaY5$iP^?ZbWfri__m6=!Z
z((El%iW)?uFv@1Q29J^d+pqpFjPn7mU%ZKTZhj9n#YW(d*}IM6-GoYtQBB$(Ygw67
zn;|AdRV^?P0lZ>@vtU`u=T!|ElvKp#S5Ee~D)(RTy^OMg>?<;vS3+nKPRWCb
z>Ccr^!h^>b^=Sw
zk%V0wB1Cv}faA$M)QlE$*4ouMwCW3}+EcKNI!x2XQhfy`hyH`(>*iZ)nCr~Ll!10U
z{!RmdARq(qh&!sB4baKpIir0PU#o0aBnpn#qbR`9WQZ_}Iggmeq-vqUF-f6r%Y#fd
zN=EvPN>ItBOp#<1jYrD+2Iku<{5vi&F@--JG~Ny6e{ph#I0ej*WrgH|swJ2%fc?=H
z{&4>nxV!lwjJk?AiBL=gRx9iH!PWOL>&|ndNQmXdshfD?>LDh77+?~0VOA_8X@K`1
z{E~$Z+RXyX^=oW;O=EWQo$tpdxU=y){PRcu6@y?KPE$6^$vDT2#n*B3;+t4(tU`y$
zpKBV$5=u%VXi6_4A|&0i$A~VXa`H@(S9m78-6Fu2y3@L~nE?mGN;5f8=AFO;A))xOM}hp${*N
zV7UeqD`g%soi`)1K|OD(^sz~kmF^x6=C<(K#dmOV{uXN59G{galV?%?7LERSo!5DN
zcj^T;fQA&c6;k{_i+I~@VQuzh_~Q&ie*#sj!mumQ^#YFLqG5Gl+cg*!2X>p?3$pY5TJ&eOKBTa$up*aR4Qbgh=F$>*$bydYu>jD-V^Bk{q
zsu7;?p5Wf-j}Rw>j47}g$+&n~d3Esc^cTiFb
z?RFE2m80t&F^>^ehI^w=;a4dBEk$k@IMNRBr|xg@bo2z7r@%^79+7LN3EkA;g(EZ?
z4%X-2#FTaopBDFVW%@cQii3zbj>Q>ok0QLXB4Tei;$4y`LgWb=j)~BlAW1ygYZVUj
z0ySZAAEt_w8&gmH=HXHy2aTwZ)A4EaUXn9v$BN`aDUlW_0x_))dzVd$KBJ~&bunkn
z!quBl=on_?$(KOQWr7QB95k)KB=fL&^c0VF?qd7o2`|8&H^i~Bi_q~=)$2&|43=TS
z)*ZN(gDg&vrwMZlN8T9sH}5jKvo>X;CaMyxq^7q>kdirdnn>QEVIWLB{OQ5(aPRmL
zM7@wBR%%BR15K+9w^N0$d+fz`LP-w8EpFHhQGt@UFqJ0SwFS&|7VtN>{uj(QW@VxV
zPftn6Be*dqBknj>5Kf2!Av1s_l2%A@k|OxCz&cx28Yes|OtO?wPvWwXQk}DxVZg1{
zq3betjUJmgV$z}g&zInaCq|ryGO2~Qb+jqqV_#Z)ixNSm{At~Qp=kWc6sO1^HB^Q)
zfsYWfx8nW15Ae~WU*ll1h0baNy^}u7pn_|wui@>>@50P9zD`r6fTK_0jcY$b@8}43
zwttVv4^dbNj-x~TY2!C=D=M_hIy&|ma!cSSI>3ide}zBZ`v;8UU6^jpzM1jRLfcrw
z+t>dFOLNPx)C!AZh}xl>60S_*ltnhCYL$_Dxol)rnGsAC=P|Z=Tlnnh`#9)rVjrJk
zl%BBABZy*FFw=EuI2H`sfo@nVCL2vW><0JHA6q;=jwe3CaLmV!u3=Kdk0*U(Mh2~F
z^QJB(&(OFi1vK)WS`-l~M$Y#cN?D3C9|z-IOhz91bj^ExT$_I#Q_dK%BKCn!5Um3Q#U{F`v?4v(*6
zRlpRE^kw5mP0$+-@#(`)@W=bVK|kGx-4ZaT0jIDa0+o54*+`&nwQynjBASgl4!ir<
z+TXz5;SjrV54B<&x~`yInMb`=m4IU|gYz}Tf}>U$0rVyw3P(fNJHSWx{t1Z{pflaV
z+)N!;W?#Zm^&-0aef(+T=NS2ih?E?KO=iX>XS*ij334w(6k90rI^uYYJexfmg|m`2
zB~$5yjh_v&8R-DE0|#niF~^a|;JFGlRe5~LZY(^@PdHYDl%@SakH^Z2P9x(SP))eC
z22^Tl70KTt1ne)5@Clt?Vg{_jVh(CTbT&9A3Zcs2Fuewcg&j&6rUohY=8-gYwzNST
zWRiW5bFab{9Zvr!i~xRGAWsd{jaj^L`Ca(Y2zPrQ
z!-of>rg4(a2OGaa1tX7a*VVdhSp;?$|kW{8u4Pjqn^Ki#>Sj6OcAtqxV
zmR*5ct0N9mOvWU(YoOMiMYAa|8ISPs<_Fl^-@((B4ZPhgzS{r*AOJ~3K~#P9Cs?1m
zg^Kwv!Qap8yw2;pP%o?jByxC>DzJ%!EIdXTXrww*Sx3uUV`Rh9Galtl$V-h~?i>h$
z5^@bi=*ze*jc&dHE8(oFjgUO{lvB-OIZsODc#D!CG!|_5Fw_xLvS*=aD%D09W)Ox8
z+o7Fz2$9n8s=^5ev}mfOQlK|J!UuQX$AjH_=&ZD1O<8z*>P@`XejCD+NP`sdV2qs9qB$u(Q#LjxE?P>twd4(`$m9}3v0sB5)Lv+O~nE#DM035riqdg_vTo)B>>|aMoy=MKBBlqtQ>ALMn8xlf;wic
z+koEScrH_=R1BY+N5W==RyKgSMBao#Bj6ZeHpbKK2l!~`a|E>jS`jk`J&R&SH)xDX
z0rg2XMy`%nXrpD#Z)^Zd7decT{7h}kTq{mpkes^WY%E>CDqLt)4IX^N5hqZZGi=C(02H;=1pH*sb4
zCFZGpQ2Yspy)j}K$c-LE-i1~RpqtX7g_?pqjya8{pd=cZrVqn$a1;&@6cbdZQ)KoS
zCxcxCn86R%0Q*1$zy2C;t^7L-TQL?_*YV)s4({%Mgx%l~CW35&JVbGX8d|8;ma)1#
zg+|?yT8Y=u06j`POeRC*
zg$hM;_%lbdQQb787*zs#c}TlAAb4~p6=g+RV!dRRU8~0W5!y;>sQ~#<;7PaPQ@(3hoxIcGL5Ip
zYjZDQk{)CK&K9=&J8-86!Ar5*dy4EqhEd=l77;cMH_(l?pw(k!N{-P<2s>WJ?d$&@
z@7?k1qTh|~mRDxeh76H>Shm_>`n`2l@SLP|GE&_D-k^(9p5
zItnsO4?KvPg35Xes#ZaqD#(NZ!_;AFa^rS5*+iUnk>ox`Ir&I_j``zR%+))n)Ye{f
zH2is;*LnT_TraKxq`pj**|NG~@qkL?f+Kxx%Pdiqg1tOMBCRCEsSgSB(v0WSg9R@z
zim6EDX~8`ZE892+j5q7j=#&~rvXr4T3)NDk&8i~BA(Z!emUf}&NHOB^9*s7mxPWe`
zIQ0A22p?f5zlWp40%_mJ0;X`Q`6CuT5K_RRgH(b7lnP>(_7^HWXElXyb}?4RoZvvw
zW;A+2X!6KRF)}_jDY^3#0+ZyJHS2hFmDGVgJ34>BI=shUX(u3`K
zIEBDw&ZaKv$TboT9Ku!-7__*D3iBcr*JO(r_TN@3NQ8re%$>8=Uu*#BJ#)_~Bq6;h
zW2i)F8h#c{m1VHlcsJs6Y!E*)j38*SG-Q4;jbKe`Ahm*{*XT7-V_nxRTwhtoV6=&k
z9}VCg_At|2!qn6vR;t&rHhUFqa}o2+1+=Oy^pkzuEj~mN453vc)=`L1N3%8sN2~M3
zf<|X#W=u&Cg1CU>Ye$zInTE#|X+NV=Oqk2(I*TJ@sRGBS!)XJZ_7AYqxqz$lx3E9n
zK+ikG{?RrLqXW2F6^pG4Sf0OtirHW^nMQkLmd*Z!)SNMoR{Bp8m7++-P>Cz1=j?5#
z(eYG0XiCTyFe)e1*nBny#R*RQEeL%KLrK^?o4f*-J6B-PxyuuNF80Rx3w0Xg$74he
z#x{BPK${EWnM^|dL7Rz8j(_PRo$}^Pu*cy>^cjBs;ooBSU=v|wh}24u3{q6H23}rz
z71x$;z%pF+^JJ?h#bqOTQsQQX0#j??>f-ki5}Kp3
zz=RU+ilhQ%QpHkr7MG{q!aG-fjA^&cO}4_ot|FU!YL0cO-e+)JSkgC-kX61X<&t}6
zfSWb?HC3I#jitAsB@`KGLKh8Gtvc#v1JWK
z{@vUC3{Q_9!6r>mK1SW?;MU6baCPw}$WR~{1$=O8O&f-7Ax})?zEmjZl>h^8h=buK
z?r;AIo6%i}EX9SXWh_t4qGD9Ok>hz@=XGA+&3bW0(5h^ZXveH5)Og5dBg-e{c$f(9
z7NdQvDJY%yIVMUgS$TQJu}l=k&t|E`7=~u@;!MHt$jb#&FZ0-Pot3x+i8oXiX`v!X
zBKGeiR3wc<=F_OI#g@K>GLcCIX@$`(%~0TF5n{{doqd{ki2Sb9uxciAsx(XGmzgDk
zxkL2AE$RZOAxs4C%1N0Q7zrwlfND4#1xF^n;E;d}ks84u-kT6`guwObt5~bev%)rx
z3+8@VrVY)^usQml7zHWzlOqHeLbn3={Rl5LR`G)?{~0q*8^iHa#99VTrz8j+$xO%u
z`Fh^&~k+F5NVo2rDo|H8bFkSU#(Of%LY0vjdlgcft{uj#Bq#Iwtt0#@l#kr;qkB<
z8Bhz88v#Ia$~aG&7h@Xdr)-34!N#T58iM6F5O-sY!#*ypuHwSNWz04%VaC1yk*mC*
za*+Tc_L0#0AesWbe%-=|U=UKyFOd$ki6Ea~XS{(ybO2EqlVSyC(#BkU0rRyr
zm?X&|NR6+DZ+Dx}&i+t==ghU0FPvy_R2_|D^GL|-WsPD0Q*8V9@!Kc=h>tgZfo9c2
zVWt?56V&21ZeRR9-o5-|Ou6&&1e3a%&Q&hCy7X_fS)+M?qj&J;#rI&S4u0{;|BGPc
zBdunvg!lSg*s2YcC?Y<^M>=R%F5#7n@8I^@yG$qGhl@#VQ@Z+*vc1sRJG5Y#m2^;K
z({i@4Ql;pK1WD{^Ev<~2`V?N7eh&*`9fe3?>UA{iMKmi-c_`Eg{G^ZWaD>sgz@?=d
zNc0RzF+fvkgZ8PW1MK@d7!CV)y8S5(hfvl5-0CbA+tXOBuS*I^Q=;{g>=;AufOD8#
zokUCyV70JhE#PwFCCn-RT15&PKNa9>3TIh0YLLuu>Mx2=N|6}h>dnCECEV1*R?dH3C?0aNHW-?b6
zS;Z={yKC)kOgAvn05e|&ebV32Zx94QfH08I0!A84&$Q50)wQzN+>*>>u8|r0?jF9{
z{)BsA&3Tu1`)!O3=1=>Se$;1
zBb%uEl#+D=J%_1--|1rEI4-(L7yVw$okrI-5%vb`bw}iS0*{2woL&03SI&`(IHP(!yJqr2h2?2NM7!5SmPSZ}0G5nV
zbS+dHCj5yQhTntR2r(g-;fomxeS&fRqMY>+Mje={1($qGSk7vn+f&2NfcEAi1*D>2
z|I|Z3z%{F*MW(A8{%Slk<74@<=UuJUyol);}kTAq6aP$f(xuc`%8B6bqHKX|j>4Erd~E9v_J~Y?I86!NMGi
zJODGDUO@otu)T|i>!0Z?Z6ZB{DXNx@@v$)m%aT1GO+9m#FSqwzzAUY*CpZ-)a6AJi
z$tK>r|1*4e>vwF~jPKgLq^HCS1REn|#Sd&kgTCElw+9YtGr
z>B4M)qjVb`YY(y-gG9{arAX`URy|gZ6L*fdkL8kwVuR=-8Q<{5NtXp58C&PLye}1X
zQ8Sm%K_qZ8nd>L?A!f43n7XwhlmI=%IR)m=kPN;-nort
z`xw4A27gLn#=ZbwRJ5Ip$a3v;#8ej&u?@nQ#YkCEumN~4YNIG>ochv63%7QDhoArM
z?^*1%Fm)NLGb^}2WvzKqZ-wpQoS+&NFYK*
zAz#dd>>#j+aTymIf)p+8JQ~gC>wI5
z!$A{+PKz5OzE{S0brKb?&ZNdHAkRM+p^6ZK6uA-q4RO^@K(T2Zl$|oh$_)&>9*+&z
zoA>dn`wm)r9rRTX+t@*;vyIir#ozqsf5e6HcQ{!i3ItnM(1oQF9N@Q~{SLS9e}ZoC
z2_j+gR@2DN$WkWTD>cAgdmWq29u7PAG4CBW#(KXGBl8ARIG-W#US
z*S0esW<6n7>Kqo7$jFG>Y#|F19(xKgfIrk*=NCa+=?fy+SG-hFmad+=<(_JK3`{@Q
z<6ni|Amp0@gSvuJo7W1FQv=j97qFbPpUZ#pQK^BdGl9}%4XGf%Hu9shA!7R!{1KLF
zMMrqJa|in;J4j8%erc90Vyrj^&nhy|XEzh=Oo)`c|MKzmR|QNoMkZbtrN}s*2For<
zn>esTUFDiOvhqAWw$4H(Fie;o-i0*VnMF_Wh^ZmYS4
zNBj2?$%sK7(eetF3Z^G#+3=iqBZ(^%88<(-$-h@|8U}SgJ&9Bfu;1CkC)>Zn?MMHN
zW`7rM!NT#NiLw~Sh59wT)_5J`)-3y5QfwP_@BEUft$$M~d5Mkm>lh3YorBhgq*9R(CxNSEl(gW_3)#We~&q_igvpXn`kbc{#@{v0k(fCn;sm?g$QgE44=EqB$|7?
ze;>cT`!l?|{!2(#;7RrvkJ>d%rNHFm64!xuWc666+9KLn9xV;#m`eifbQ1@?Ck#?)9vb!&$Z`>e^cgi6mP9x1;_=}c9v^&$X1Ieu1ZcHS5QLF}6;cxE+3qC%gBtcJKk_i$BHE_+?BqZeYB88RLZs
zOw^_^UT8oN2PiG-ZjVawg}H4xt}?6?jz>ufWcks9u-g=Bl}l7LSY*>}X(MFm!Al$y
z=l`o4TW7C1{5ffcXrJa6tKmh8_tY~lqAzH=<3&-V
z3{)Fy2Or_H%@45K*+8JE<4jgW4J=GtM#XLDyw`J0M*5vb>}+Q*pVlzM1Lt2Dse#F5bTQBV3rff~q?SLs`soHS!5FStWJi1N6Fm9JN|_vi}hG_ddfm
zeMUta!XZl^15s$;c#8x)6iwKG=qbR;U`Z3EPaI=~X0(rwwtmIdGc~_~<>||)SLRVI
z%%JGiQ1D9ZO5l_8qGQ4(q?hJYWlJARM!w5AD;OyP)km-2L%%)1(O?5>yZ7;Me+`Gh
z9wM`Y)a+qkbP>x4t@aT<{P(=lp6r^~Y7OP6R^ZJYhlb*)l~dYU0t+U5;Z4j5Y?8yG#CnrUidsohDK6R}KgC
zX^G&YgfCr;Pmr3;=aX5IUyIE=xSc?e$|9JC%_AF9AD0W1Hku;roxmdHs%m5V_yINw
z)2O&($aJC6YfMX1r9Xxt`3w!N%
z@woXeo*X?uCsQbw=Wu!U1{RiH$Nu0L!FrD+qR}8gK+$9(MUMjJBZ2z|x3M=^N3D4e
zMQ;){e*%lsE4VQE8m9b3xN-uPD#Le;Jkj-wNXO^gsWS1C{5p>#O%Oar?
zRtFmxCR3yk2HS+=?%s!ZxPJ?oHGpmVI5|o&;ZI>{Yy}NxOlR<>EHmL7rLVgTL59^K
zltG!;fXPWY1CFkvn+8{U7;BU{Yx&yJ8}JJi9vdeFst~ZvOS83u!}dDbodawfJ;BED
z24{Jnq%Fio0Mik$GaFXsz*iO2yfG|JT*A`qHS8U4W9N7i2jl~pp766~^t__Qoz~-M
z3!C@up;DYerAUU{6DSvJXjH1G`bCuOGD=RBg*?Ti%5)Q=eGKA=1sf;99@_B^+Tjse
z-8PyhUF;p+!*OQ^Ju)I!eHNc(nSv_|n5?W~xv_$o#w^CmlPowfy`d>zf?j;grqaUD
zG>kM(k=lyQy`wB-A(SlzrfQRT!*~me=`7Pj_I7sBZ0_Np7oZ>PVB@%pLg_xn4#zQf
zG=r(~4C>}IrYozMt1Uy4Z{{$XWmKaY)gOJmG5|;*ELPg6mlMI(U4~JVD7Zz~
zu7yq%;;7fg{k;!y*n5Qi?I(D+_FJ6v*Abx$X%|o{&tST78Go_*H&~dxj`ue{!rJC-
z1j#Yy-ug}jD|1&dJ-LLdgG+dPa0{*80Xp#jyT>0vwj$g=DPVka4l|`|n6JHsx$!r#
z>P&-Ee4ibs4l5tYrlEl(a@pvb#`Rmfdq|TH5C(}>r&1TzNMIxqZfJdZSCIqsV-3Lk)ovnAV-MxoidW<9zzlqwf1Z5wjRMQD()&Gv@kWbgoXMFb2tfd5;sp#
zma_fzpCU}DK!=F7#~qV9AUfl7`HZlg|%byFx*Fw#q3iUcMi#k2*pYfRf_&dXd8ZRFtC|lD18(Cq{sWHTCU(``Au9%FXCiy
zjN{f0TJA2^3mdr8eFP~xh{^FZ-a`NA0ruM_ti(f6&Eewg+xX6vpWtHcB5Y}oI)9@Y
z)t^YcLKUz)6`fJ1!InP99u;K`5ydqn3SLx%ivpzQpx_lzD%DUaHc;}aaGf$dw}gV%
zfbUlr(7B{+wW*WkGFUK9gE}*vbIK+$CbB4nZAk`RzEMFh=)#1@4*e%b53vzHuJjt_Eb=GRpFX7VsRbHz&JHnIJJ9x7H7@NCyuyynadeJtT
zojq*tuHm70fa@39@M;z=$P$1tyc-lXja;
zgiV+DT?D`)*kq6KrKPyDCxke)d>H~j%d!}}4S(i4le|x2@=+67*zjV&D)Me|cA6;6
zCgQ~Ci%wZDVgk3YfI@dNa6j4TV04FqcL1m2u`i)}~>vZTG=vPgG*
zbGw|Bb1kG6&5AGUpq|aN*?x$-LBt=lm`gV7crxHqx>H2!ZSq*y=IFs72yk!nHr~Jc
z4+upMj#q&YRrJ$6I2DCrrHtL}Bi?@U&?K_|AZg>{Pd>!`^*cNwWmmdDp<<6=seTbt
z#W|Rnr8%Ct)>XdzF#Dy8R;O6-*rF$jdu-r5H+C5{_dW`7if>>10WQp}pkd7*NJBi_
zyN}=A{w1F5-$y_>$7Y0H&_OTgp=cLSm@L9{O|%a>2--1hyMU?65*8L$IJ0@NzKq4O
zWt5BtTZT>*7BO{Y4%c7%7M`3u!u{<}84PY5-pA3_K7^5=?pJsMKsm`-HsCDZR+u2}
z#PC!FljSD!a$lePHXKX)F?xm#w?LkRJ?u5tvESc-^jt`%2;VQFVA?1e9?D(=i&K|z
zb?I#^7jM8;K6Vc_@&Ep(|BR-2fWnLqD>B&TD9E}vig$66Hd$RyHa_%t->GsAa3=K4
z4;k1yss!7ppkAE9lr@jkZDDn)iP?A$o4b$kWcO3-bv{Fy_Fy;(!t?~KqY&*TDUm!}
z__GV_dO+RZ(@#+TdBg}1j%rkYboJ^0AkXO;}Cm3Bda_bmM+hfBTNDL1jTqd=cDe;
zV63==W_N&YGvw$lQf?Lt7T+y*DK0MYQEEq(q6@<+r
z9PEFJX8Q<9Jcr533a-w)jcW_v#-*7hRHjfe`TYD2{8_36fss`
z!LoN98>P$G8h;HZ(G%=7?_+yw4Tx=2{Sqe2RZb}RvI#JO6x+*SR#T(_ap+E1_@Y9I
z-r$6}l_seorQ%qxz#QO7-cm1$Xv9)q$Ow`A}}7pGaPrn`g;5G7xr^w)V1
zs5D`47|?GRxwt1nCSqjdrI*I&_j>qf>$muL=QlVR1UT#*qnCD&WEm{wV7xqwtFzz5
z_m+N$1+s}5J~gy*mk=M_RHW=XsmIhpyd-22WaeDdb7%rxO6swz5P8Qm`hc>8OFt-V!oer$j{M)iUJ}$
zVsT;x7w4{F%$sI03~@ni%Rpopd}H%<4}s$-VD1{J?Hy9}B16ra$K~1Yz_ESYn0*H&
zvj_`5gZzZ{Q*?;q!8#JRheFXq!8ITUMIOVp4_eH%o31Zneqk9Gr>|gnd=+DbY51}X
z+i+piRb_7b^u!6&7%H@!ME&+CAa;6SKH5H-pLY1nSN_O4ba&M~asW(jEfJ?Csq_vX{V1JT{6C<31vz$CfzE_KYl;
ziG#8?hDu=^MbqVTP7=gO2sAndOvi%n)IqT!o*F}4Ttsi`CbpWd4BN#i3GtRHGWzA9uY%5l9A;tSq98fluA3MqRLsf$j|eHbPs4cMIe5v2kaC1{pMez
zvNwZAhmY`N?-63z2I?7Foqc4Y4a+L?XgB6W1DnS|3zPG_t#3{L5Op;MgbJpSvVRiG$D1>Mh1PZxd;U7AV>4J<
zcn9;w1-^Atm
zMYNI@c3O|IUUtT3J8r;Wgg+hfX&oALu3e5jGco$e6;q@cyIp~fG9w^E(;CPP@v+C;oFyf
zh__~ch=syyFasNjsbCIOnGDj*{02aA8i*~7b0kB7C=Jk$+sNqnAjwi3Am;9&end4efwgzi!gEo9iTMUR+GHqCUHcS&h&WX|l!3ha^(QB??%J_d$F8UYQ@gX&9cZ0=#A
zI*;$a`B!*t=1m?sx=szNm#*NTu#bCtyJ#Nn0eu%M^OvwVz6z6kK$x2ma$avL^g!GF
zntOn(ZYK9<%GY5*kHI%$Z(Xmzw;YU_)BIV=n8WK}W^4*87iY0GcpsbHbqsnY9IMJn
zJ+UleY5WS7OG{Xoynw~VD(c=eibhGtwdH5HFjEK+!jci&>JWuQW(Hhk!!=82SkstY
zT*mA5HjX+wxZC**yGPsDJbHlUU;|my!64})?qsNmd5jkqaG`!37TG9;F|yQw?MhVa
z2CmNEMA+TMj6aQfV+jijH&FMM(C}-h`aZnO)_VCsfByL%u+~F3zKxx&&(I5F)Xge)
zlVcKD(fzDt^8Tf`#QE7pTwJ<>x>JF{X8DRasTA8s5p?9OKtdqfh~Y|!a^z_DfvE}1
zmshbp_vbk3u48@s9!^>Qgs1>d`i!${m*y)Mu2csqx$2kSD5`Hbf?469&gkU
zunG-;zJ}+T>;>=Yz
zr2@**!kx~$s8yG-GJctb4P2CJlIcBfQbzDMaNTENNv8MDP#sitYp8lPeaxM^=ODFq
zqd19`nVZ<&+eQ}Iybj}qIgFKNQ7KH|(!@1f9KVEuT_Wb3{=+c8F&r*o4r~Cz%x6#0fOHFR2t
zWKU}oi}lywUvbe}XrojZN29R-M}KPbZBu7&l5Z;w6hs-zW7n|LKgMy^!bE{`&uuMI
zGEMCVsT@?CFjLXy4;83id(kDr28LX;K^HU4^UaI`f>!=mxQBM{oF;ByJy%F#m)u{e>>lFe(2_+r2p+Mg4;Tp<7an|rV^?Cck5-%{a
zC%-ZO^#6=J5so}@hvlg<7R)jhrdEN;bDc$T4!E5sLJj{rJb$~$0gA=eFzS4_Z1_Oy
z+~3d>xaT&J5I`ru)UcPk|%R!|+b+#$Qu}d@$kHw#qXpq^tsTp}*3}8fkEPJoP
znDWueIxsR5wQ>V1#fzA?=i$q`KCB2iSonI<1nK_dMC`(qC5+Xl(Qqb^
z5~y;ti7^Gd(0h1x8p}5wIVzxW_u_OJUj4s^Uv#i`Zxg*{4;Mw9DL25nWP#fUM&Ii
zYo%|aF<-}Iasl%*%RIJZ!L*@}(F+(_2;pQ;gXgcSXpAlM^%{Qe`G3`rW2$(0jp4jK
zvr&^PczyccV#<6Aj_F}>ashK=^Qichr$~xwcr!TsjOE7;MFVOO=AJz!4>>u-+e+)X
ziTzj>6*S5XTrOYZ{`}tI0gjH2IHS2zD`R53!M=k8I6Zr))X_8qN0czvxB)cIwkt^n
z4Ep%^@p0DOhQS8o)&3m5Tlo*DEv@1>+h#7XV-w}Y=I$PKElX}0)k=-|y+c4r-SP9E
zv0(-ajnaqfp?poZ8J@Ffxr)L;vAP7%SDv0DMoA!}8r7dvz2fw5^?YOKn=l@Grm~^I
z@QW%BHC5+9D0d}ta=sdNY|aguU-(?=h2K8=tKkox{gKGU4bRp26bS++DT!HXxG;AO
zwebe7ge@KgHQX^&j0qTYwV`-2p`KFWs=Q0^>Jq`zegMZR4eZONro)?y#NvFaG=ho|rHMwN26xrM!V;y~
zSSVR#xCDaO98nCh<5zqB&n?CNkl>NL{+sbe9o1@yWA~`*M!<`{Co3B|Cp@;)`ElBC*##Wcm2dL*9i2g<(AmC~ju
zd5me3^RqE*cL{i|?m`K9t^*C6j?y)V&EQx*>ee*xGd*I~Tr%2!QVzE%!%J=4H&i^v
zSM2u0JYiq>@NDXq8#zESIc0;%6V_M2cVAFL;SmY7hT^BE!qb;A5(qFc8X^%W+cjB
zTNb4+$ON`bQeF?OxX(3*vg~YPelEWE_j`*k*j^*?c@&ycmbmEsrg;d5RujMcYm3Z;
zommXn@)We3gQz=iv<9b?fS0V($tib-1jpCqZ%F?KUkzJ{d@W?2&Y>-T=s3>f3NuX`
zASI{F!JjrxSEs;}fnEM4SV2$6OE&EC@*|kjHcVo9zODB4)Qk4%iz_#LCt!~hg2A-Z
zN^8i|wLN8sd<~ncm@O=G-f+HtX)cb)Q)LX!3s($8pGu>x%;&$>`5JNfMm4HYeO~=Zb^yO@l=q?$&hSL~>{t@7
z@+3>w0Vf|5D}C9@a=M;m@3{N|;;0;eyxevwHOx*Pf(gL@Fimrz3U%Ew1({NQt(AiX
zbv9wnmsg45==O`(`zu%a40w`5yqTSXVvT`b?sX|ImpE`LJJnq1wdM8I;BRO%^D_Hd
zYnz6sMC$I^kUGHX_I$^D1%S-YDKtPE`XfH2h*u)tMv>EK75fJc7ic)zAM)QAc+lvS
z#+gcQlVRkYzx>ZM=25SV8#e4>8kvzBehZzqJXHE;LuL1KfBG+f&7s$2riD4D`Yn~6
zCe%<2p;l~vg@rjC6XyV&{6X`ZtKlmv>hO}-hu&8%#udgh;FkVt=N-UNWXGsRHLBmE
z{v-k5i|Ra}dl3M9w*GY}<#M?rfeU4E4~IdEnYRwOO%&&(E(fRPh+kMji$S=Xq`?t}
zT7aC0#uvmaL6#C>0=(I&bXy>^}cGM!qj9O?$|=^E*5YjC{>irt%J~$~k;5
zc!C#{G>m69T0J$1ptUAWA$@=iTR-XW$j$PhTD2>xbGj}%W28Bk!Lkh&#S^9=Rb3WCO
z205JIOL4j#3NPctJTLj160YV7Cb|hqlZ1c7
zQGC+Kfm^OL`ig6GHimz}@y)g{>Z#i;g<*%*95N!$I0~a0)u=}Gd(@vK06aAp{(VlP
z=keJ$axVT7TKSOYl{!xecc~l5AW-V7qY?&Ra9;X4ndJD*#Azo_<#r-N5Gk~(g!6Bc
zOypxpa8E*G8^xjwG3=Op;Tw78Z4KYY7vNu&^9$9`v{uR7<>qY7jIVNylmdMQhfryj$Llj$S)oSH;&H)CWUn+Drj*m5|d6O?KXF1pPljq#E&teSqzLY9wI6fsVG0r2xGw$ffwPjSJ
z8rAPnuLuCDTp0-_$n&V|^e#UiG0LGOGPJi`-uO9x^PA-sB~K+N4@HsN_PR-D-CA~
zRDKI;nDwip--i2z>{GY$vV=R&_!@GNBaOO?tp(?BI#4VPnDmWax10b&$vba
za8#oj)z@5qZ~!<|n3A1UoGHW+-BFuxZNZ$RBvw3