From 58f4a222caaadd5408f95dea7e09e6e30f12357a Mon Sep 17 00:00:00 2001
From: an1018 <614803115@qq.com>
Date: Wed, 13 Oct 2021 11:58:46 +0800
Subject: [PATCH 1/2] add fireDetection
---
examples/Fire_Smoke_Detection | 1 +
1 file changed, 1 insertion(+)
create mode 160000 examples/Fire_Smoke_Detection
diff --git a/examples/Fire_Smoke_Detection b/examples/Fire_Smoke_Detection
new file mode 160000
index 000000000..2c204d64d
--- /dev/null
+++ b/examples/Fire_Smoke_Detection
@@ -0,0 +1 @@
+Subproject commit 2c204d64db17e231c305d4faaaa5c85ddc7c72a7
From d85ea35ac894fd76cd5ea892d644187e0df70b7d Mon Sep 17 00:00:00 2001
From: an1018 <614803115@qq.com>
Date: Wed, 13 Oct 2021 13:07:31 +0800
Subject: [PATCH 2/2] add firedetection
---
examples/Fire_Smoke_Detection | 1 -
examples/Fire_Smoke_Detection/README.md | 243 ++++++++++++++++++
.../accuracy_improvement_frcnn.md | 43 ++++
examples/Fire_Smoke_Detection/code/infer.py | 10 +
examples/Fire_Smoke_Detection/code/metric.py | 115 +++++++++
examples/Fire_Smoke_Detection/code/predict.py | 21 ++
examples/Fire_Smoke_Detection/code/train.py | 55 ++++
.../docs/images/dataset.png | Bin 0 -> 715345 bytes
.../Fire_Smoke_Detection/docs/images/demo.png | Bin 0 -> 1394928 bytes
.../docs/images/predict.jpg | Bin 0 -> 52138 bytes
.../docs/images/predict_result.png | Bin 0 -> 290108 bytes
11 files changed, 487 insertions(+), 1 deletion(-)
delete mode 160000 examples/Fire_Smoke_Detection
create mode 100644 examples/Fire_Smoke_Detection/README.md
create mode 100644 examples/Fire_Smoke_Detection/accuracy_improvement_frcnn.md
create mode 100644 examples/Fire_Smoke_Detection/code/infer.py
create mode 100644 examples/Fire_Smoke_Detection/code/metric.py
create mode 100644 examples/Fire_Smoke_Detection/code/predict.py
create mode 100644 examples/Fire_Smoke_Detection/code/train.py
create mode 100644 examples/Fire_Smoke_Detection/docs/images/dataset.png
create mode 100644 examples/Fire_Smoke_Detection/docs/images/demo.png
create mode 100644 examples/Fire_Smoke_Detection/docs/images/predict.jpg
create mode 100644 examples/Fire_Smoke_Detection/docs/images/predict_result.png
diff --git a/examples/Fire_Smoke_Detection b/examples/Fire_Smoke_Detection
deleted file mode 160000
index 2c204d64d..000000000
--- a/examples/Fire_Smoke_Detection
+++ /dev/null
@@ -1 +0,0 @@
-Subproject commit 2c204d64db17e231c305d4faaaa5c85ddc7c72a7
diff --git a/examples/Fire_Smoke_Detection/README.md b/examples/Fire_Smoke_Detection/README.md
new file mode 100644
index 000000000..4e2d104f3
--- /dev/null
+++ b/examples/Fire_Smoke_Detection/README.md
@@ -0,0 +1,243 @@
+# 火灾烟雾检测
+
+## 内容
+
+* [项目说明](#项目说明)
+* [环境要求](#环境要求)
+* [下载PaddleX源码](#下载PaddleX源码)
+* [数据准备](#数据准备)
+* [模型选择](#模型选择)
+* [模型训练](#模型训练)
+* [模型优化](#模型优化)
+* [模型评估](#模型评估)
+* [模型预测](#模型预测)
+* [模型导出](#模型导出)
+* [模型部署](#模型部署)
+
+
+
+## 1 项目说明
+
+本案例基于PaddleX实现对烟雾和火灾的检测，主要针对城市和森林应用场景，希望通过梳理优化模型精度和性能的思路能帮助用户更高效地解决实际火灾和烟雾检测应用中的问题。开放烟雾和火灾数据和预训练模型，并提供服务器Serving和Jetson NX边缘侧芯片的部署指南。
+
+
+
+计算机视觉中的烟雾和火灾检测在住宅、加油站、公路、森林等室内外场景自动标注疑似烟雾、火灾，实现实时自动烟火检测，具有重要意义。
+
+
+
+## 2 环境要求
+
+* PaddlePaddle >= 2.1.0
+* Python >= 3.5
+* PaddleX >= 2.0.0
+
+安装的相关问题参考[PaddleX安装](https://paddlex.readthedocs.io/zh_CN/release-1.3/install.html#pip)
+
+
+
+## 3 下载PaddleX源码
+
+```
+git clone https://github.com/PaddlePaddle/PaddleX -b develop
+cd PaddleX
+git checkout develop
+```
+
+
+
+## 4 数据准备
+
+---
+
+本案例使用数据集包含互联网下载图片和已公开数据，共有6675张图片，分别检测烟雾和火灾，示例图片如图所示：
+
+
+
+数据集图片格式是VOC数据格式，VOC数据是每个图像文件对应一个同名的xml文件，xml文件内包含对应图片的基本信息，比如文件名、来源、图像尺寸以及图像中包含的物体区域信息和类别信息等。
+
+xml文件中包含以下字段：
+
+- filename，表示图像名称。
+
+- size，表示图像尺寸。包括：图像宽度、图像高度、图像深度。
+
+ ```
+
+ 500
+ 375
+ 3
+
+ ```
+
+- object字段，表示每个物体。包括:
+
+ | 标签 | 说明 |
+ | :----------: | :----------------------------------------------------------: |
+ | name | 物体类别名称 |
+ | pose | 关于目标物体姿态描述（非必须字段） |
+ | truncated | 如果物体的遮挡超过15-20％并且位于边界框之外，请标记为`truncated`（非必须字段） |
+ | difficult | 难以识别的物体标记为`difficult`（非必须字段） |
+ | bndbox子标签 | (xmin,ymin) 左上角坐标，(xmax,ymax) 右下角坐标， |
+
+使用PaddleX 命令即可将这6675张图片按9:1比例随机切分，切分后包含6008张图片的训练集和667张图片的验证集。包含以下文件夹和文件：`images`,`annotations`,`labels.txt`, `train_list.txt`和 `val_list.txt`，分别图片、xml标注文件、存储类别信息、训练样本列表、验证样本列表。训练样本列表和验证样本列表的每一行格式为：`图片路径 对应的xml路径`，例如`images/fire_00348.jpg annotations/fire_00348.xml`。
+
+最终数据集文件组织结构为：
+
+```
+ ├── dataset
+ ├── annotations
+ │ ├── fire_000001.xml
+ │ ├── fire_000002.xml
+ │ ├── fire_000003.xml
+ │ | ...
+ ├── images
+ │ ├── fire_000001.jpg
+ │ ├── fire_000003.jpg
+ │ ├── fire_000003.jpg
+ │ | ...
+ ├── label_list.txt
+ ├── train.txt
+ └── valid.txt
+```
+
+
+
+## 5 模型选择
+
+PaddleX提供了5种目标检测模型：FasterRCNN、YOLOV3、PPYOVO、PPYOLOv2和PPYOLOTiny。
+
+* [Faster RCNN](https://arxiv.org/pdf/1506.01497.pdf)：Ross B. Girshick在2016年提出了新的FasterRCNN，需要先产生候选区域，再对RoI做分类和位置坐标的预测，这类算法被称为两阶段目标检测算法。在结构上，Faster RCNN已经将特征抽取(feature extraction)，proposal提取，bounding box regression(rect refine)，classification都整合在了一个网络中，使得综合性能有较大提高，在检测速度方面尤为明显。
+* [YOLOv3](https://pjreddie.com/media/files/papers/YOLOv3.pdf)：Joseph Redmon等人在2015年提出YOLO（You Only Look Once，YOLO）算法，通常也被称为YOLOv1；2016年，他们对算法进行改进，又提出YOLOv2版本；2018年发展出YOLOV3版本。YOLOv3使用单个网络结构，在产生候选区域的同时即可预测出物体类别和位置，这类算法被称为单阶段目标检测算法。另外，YOLOv3算法产生的预测框数目比Faster R-CNN少很多。Faster R-CNN中每个真实框可能对应多个标签为正的候选区域，而YOLOv3里面每个真实框只对应一个正的候选区域。这些特性使得YOLOv3算法具有更快的速度，能到达实时响应的水平。
+* [PP-YOLO](https://arxiv.org/abs/2007.12099)：PPYOLO是[PaddleDetection]()优化和改进的YOLOv3的模型，其精度(COCO数据集mAP)和推理速度均优于[YOLOv4](https://arxiv.org/abs/2004.10934)模型。PP-YOLO在[COCO](http://cocodataset.org/) test-dev2017数据集上精度达到45.9%，在单卡V100上FP32推理速度为72.9 FPS, V100上开启TensorRT下FP16推理速度为155.6 FPS。PP-YOLO从如下方面优化和提升YOLOv3模型的精度和速度：
+ - 更优的骨干网络: ResNet50vd-DCN
+ - 更大的训练batch size: 8 GPUs，每GPU batch_size=24，对应调整学习率和迭代轮数
+ - [Drop Block](https://arxiv.org/abs/1810.12890)
+ - [Exponential Moving Average](https://www.investopedia.com/terms/e/ema.asp)
+ - [IoU Loss](https://arxiv.org/pdf/1902.09630.pdf)
+ - [Grid Sensitive](https://arxiv.org/abs/2004.10934)
+ - [Matrix NMS](https://arxiv.org/pdf/2003.10152.pdf)
+ - [CoordConv](https://arxiv.org/abs/1807.03247)
+ - [Spatial Pyramid Pooling](https://arxiv.org/abs/1406.4729)
+ - 更优的预训练模型
+
+* [PPYOLOv2](https://arxiv.org/abs/2104.10419)：相较20年发布的PP-YOLO，PPYOLOv2版本在COCO 2017 test-dev上的精度提升了3.6个百分点，由45.9%提升到了49.5%；在640*640的输入尺寸下，FPS达到68.9FPS。 PP-YOLOv2在同等速度下，精度超越YOLOv5，主要改进点：
+ * Path Aggregation Network
+ * Mish Activation Function
+ * Larger Input Size
+ * IoU Aware Branch
+* PPYOLOTiny：在当前移动互联网、物联网、车联网等行业迅猛发展的背景下，边缘设备上直接部署目标检测的需求越来越旺盛。生产线上往往需要在极低硬件成本的硬件例如树莓派、FPGA、K210 等芯片上部署目标检测算法。而我们常用的手机 App，也很难直接在终端采用超过 6M 的深度学习算法。如何在尽量不损失精度的前提下，获得体积更小、运算速度更快的算法呢？得益于 PaddleSlim 飞桨模型压缩工具的能力，体积仅为 1.3M 的 PP-YOLO Tiny 诞生了！PP-YOLO Tiny 沿用了 PP-YOLO 系列模型的 spp，iou loss, drop block, mixup, sync bn 等优化方法，并进一步采用了针对移动端的优化策略：
+ * 更适用于移动端的骨干网络：MobileNetV3
+ * 更适用移动端的检测头（head）：深度可分离卷积
+ * 去除对模型体积、速度有显著影响的优化策略：iou aware、matrix NMS等
+ * 使用更小的输入尺寸
+ * 召回率优化
+ * 更大的batch size
+ * 量化后压缩
+
+根据部署场景，这里我们选择了双阶段的Faster RCNN算法进行火灾和烟雾检测。
+
+
+
+## 6 模型训练
+
+PaddleX提供了单卡/多卡训练模型，满足用户多种训练需求
+
+* GPU单卡训练
+
+```
+export CUDA_VISIBLE_DEVICES=0 #windows和Mac下不需要执行该命令
+python train.py
+```
+
+* GPU多卡训练，例如使用2张卡时执行：
+
+```
+export CUDA_VISIBLE_DEVICES=0,1 #windows和Mac下不需要执行该命令
+python -m paddle.distributed.launch --gpus 0,1 train.py
+```
+
+
+
+## 7 模型优化
+
+### 7.1 精度优化
+
+本小节侧重展示在模型迭代过程中优化精度的思路，在本案例中，有些优化策略获得了精度收益，而有些没有。在其他质检场景中，可根据实际情况尝试这些优化策略，具体请参考[模型优化文档](./accuracy_improvement_frcnn.md)
+
+采用PaddleX在Tesla V100上测试模型的推理时间（输入数据拷贝至GPU的时间、计算时间、数据拷贝至CPU的时间），推理时间如下表所示：（**20次推理取平均耗时**）
+
+由于参与推理的图片较大，因此预处理时间在**Resize处**会消耗较多的时间。
+
+| 模型 | 推理时间（FPS） | Recall | Error Rate |
+| --------------------------------- | --------------- | ------ | ---------- |
+| PPYOLOV2+ResNet50_vd_dcn+SPP=True | 22 | 96.3 | 11.1 |
+| PPYOLO+ResNet50_vd_dcn | 20 | 90.0 | 8.81 |
+| YOLOV3+DarkNet53+img_size(640) | 21 | 88.4 | 6.01 |
+| FasterRCNN+ResNet50 | 17 | 96.7 | 8.1 |
+| FasterRCNN+ResNet50+DCN | 17 | 94.5 | 7.6 |
+
+注意:
+
+- **608**的图像大小，一般使用默认的anchors进行训练和推理即可。
+
+**优化进展说明**：
+
+- 1.通过选择更好的检测架构可以提高检测的Recall值——即**Neck，Head部分的优化**可以提高Recall，但是Error Rate也有了一定的增长。<**YOLOV3 到 PPYOLOV2**>
+- 2.两阶段算法相较于单阶段算法在检测精度上更加准确，Recall和Error Rate都有了显著提升，但是推理时间也会相应增加。<**PPYOLOv2到FasterRCNN**>
+- 3.通过FasterRCNN的两个实验可以看到，**加入可变形卷积DCN后，Error Rate可以得到微小的改善，但是Recall下降明显**。因此综合考虑后采用不添加DCN的模型。
+
+通过以上的简单优化方式，获取了两个较好的模型结果：【前者模型**速度更快**】
+
+| 模型 | 推理时间（FPS） | Recall | Error Rate |
+| --------------------------------- | --------------- | ------ | ---------- |
+| PPYOLOV2+ResNet50_vd_dcn+SPP=True | 22 | 96.3 | 11.1 |
+| FasterRCNN | 17 | 96.7 | 8.1 |
+
+
+
+## 8 模型评估
+
+模型迭代完成后，计算不同置信度阈值下图片级召回率和图片级误检率，找到符合要求的召回率和误检率，对应的置信度阈值用于后续模型预测阶段。
+
+```
+python eval.py
+```
+
+【名词解释】
+
+* 图片级别的召回率：只要在有目标的图片上检测出目标（不论框的个数），该图片被认为召回。批量有目标图片中被召回图片所占的比例，即为图片级别的召回率。
+* 图片级别的误检率：只要在无目标的图片上检测出目标（不论框的个数），该图片被认为误检。批量无目标图片中被误检图片所占的比例，即为图片级别的误检率。
+
+
+
+## 9 模型预测
+
+加载训练好的模型，置信度阈值设置为0.5，对验证集或测试集图片进行预测：
+
+```
+python predict.py
+```
+
+可视化预测结果示例如下：
+
+
+
+
+
+## 10 模型导出
+
+在模型训练过程中保存的模型文件是包含前向预测和反向传播的过程，在实际的工业部署则不需要反向传播，因此需要将模型进行导成部署需要的模型格式。 执行下面命令，即可导出模型
+
+```
+paddlex --export_inference --model_dir=./output/deeplabv3p_r50vd/best_model/ --save_dir=./inference_model
+```
+
+预测模型会导出到`inference_model/`目录下，包括`model.pdmodel`、`model.pdiparams`、`model.pdiparams.info`、`model.yml`和`pipeline.yml`五个文件，分别表示模型的网络结构、模型权重、模型权重名称、模型的配置文件（包括数据预处理参数等）和可用于[PaddleX Manufacture SDK](https://github.com/PaddlePaddle/PaddleX/tree/release/2.0.0/deploy/cpp/docs/manufacture_sdk)的流程配置文件。
+
+* 更多关于模型导出的文档，请参考[模型导出文档](https://github.com/PaddlePaddle/PaddleX/blob/release/2.0.0/docs/apis/export_model.md)
+
+
+
+## 11 模型部署
+
diff --git a/examples/Fire_Smoke_Detection/accuracy_improvement_frcnn.md b/examples/Fire_Smoke_Detection/accuracy_improvement_frcnn.md
new file mode 100644
index 000000000..3ac9aa7be
--- /dev/null
+++ b/examples/Fire_Smoke_Detection/accuracy_improvement_frcnn.md
@@ -0,0 +1,43 @@
+# 精度优化思路分析
+
+本小节侧重展示在模型迭代过程中优化精度的思路，在本案例中，有些优化策略获得了精度收益，而有些没有。在其他场景中，可根据实际情况尝试这些优化策略。
+
+## (1) 基线模型选择
+
+相较于单阶段检测模型，二阶段检测模型的精度更高但是速度更慢。考虑到是部署到GPU端，本案例选择二阶段检测模型FasterRCNN作为基线模型，其骨干网络选择ResNet50_vd。训练完成后，模型在验证集上图片级别的召回率为96.7%，图片级别的误检率为8.1%。
+
+【名词解释】
+
+* 图片级别的召回率：只要在有目标的图片上检测出目标（不论框的个数），该图片被认为召回。批量有目标图片中被召回图片所占的比例，即为图片级别的召回率。
+* 图片级别的误检率：只要在无目标的图片上检测出目标（不论框的个数），该图片被认为误检。批量无目标图片中被误检图片所占的比例，即为图片级别的误检率。
+
+## (2) 基线模型效果分析与优化
+
+| 模型 | Recall（图片级别的召回率） | Error Rate（图片级别的误检率） |
+| ------------------- | -------------------------- | ------------------------------ |
+| FasterRCNN+ResNet50 | 96.7% | 8.1% |
+
+由于烟雾和火灾的特殊性，无法精准的计算mAP，因此本案例采用图片级别的召回率和图片级别的误检率作为最终指标。
+
+从分析表格中可以看出，对于没有烟雾和火灾的负样本，存在较严重的误检情况。因此，将验证集上的预测结果进行了可视化，发现由于自然图像中的火焰和烟雾形状非常不规则，导致网络很难提取到非常准确的特征，很多预测框位置并不准确，同时容易将负样本或背景中与火灾烟雾类似的目标误检测成火灾或烟雾。为了解决该问题，选择在骨干网络ResNet50_vd中使用可变形卷积（DCN）。重新训练后，图片级别的误检率降为7.6%，但漏检有了一定增加，图片级别的召回率变为94.5%。
+
+## (3) 数据增强选择
+
+| 训练预处理 | 验证预处理 |
+| ------------------------------------------------------------ | ------------------------------------------------------------ |
+| RandomResizeByShort(short_sizes=[640, 672, 704, 736, 768, 800], max_size=1333, interp='CUBIC') | RandomResizeByShort(short_sizes=800,max_size=1333, interp='CUBIC') |
+| RandomHorizontalFlip() | Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]) |
+| Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]) | |
+
+在加入了[RandomResizeByShort](https://github.com/PaddlePaddle/PaddleX/blob/dc14077a3b0e8ab9e512a78938d8f01b969f0942/paddlex/cv/transforms/operators.py#L394)、[RandomHorizontalFlip](https://paddlex.readthedocs.io/zh_CN/develop/apis/transforms/det_transforms.html#randomhorizontalflip)这两种数据增强方法后，对模型的优化是有一定的积极作用了，在取消这些预处理后，模型性能会有一定的下降。
+
+**PS**：建议在训练初期都加上这些预处理方法，到后期模型超参数以及相关结构确定最优之后，再进行数据方面的再优化: 比如数据清洗，数据预处理方法筛选等。
+
+## (4) 背景图片加入
+
+本案例将数据集中提供的背景图片按9:1切分成了1116张、135张两部分，并使用(3)中训练好的模型在135张背景图片上进行测试，发现图片级误检率高达21.5%。为了降低模型的误检率，使用[paddlex.datasets.VOCDetection.add_negative_samples](https://paddlex.readthedocs.io/zh_CN/develop/apis/datasets.html#add-negative-samples)接口将1116张背景图片加入到原本的训练集中，重新训练后图片级误检率降低至4%。为了不让训练被背景图片主导，本案例通过将`train_list.txt`中的文件路径多写了一遍，从而增加有目标图片的占比。
+
+| 模型 | 图片级召回率 | 图片级误检率 |
+| ----------------------------- | ------------ | ------------ |
+| FasterRCNN+ResNet50 | | |
+| FasterRCNN+ResNet50+ 背景图片 | | |
\ No newline at end of file
diff --git a/examples/Fire_Smoke_Detection/code/infer.py b/examples/Fire_Smoke_Detection/code/infer.py
new file mode 100644
index 000000000..b34a25c04
--- /dev/null
+++ b/examples/Fire_Smoke_Detection/code/infer.py
@@ -0,0 +1,10 @@
+# coding:utf-8
+import os
+os.environ['CUDA_VISIBLE_DEVICES'] = '0'
+
+import paddlex as pdx
+predictor = pdx.deploy.Predictor(model_dir='frcnn_dcn_inference_model/inference_model',
+ use_gpu=True)
+result = predictor.predict(img_file='test_imgs/fire_1.jpg',
+ warmup_iters=100,
+ repeats=200)
diff --git a/examples/Fire_Smoke_Detection/code/metric.py b/examples/Fire_Smoke_Detection/code/metric.py
new file mode 100644
index 000000000..453e12b02
--- /dev/null
+++ b/examples/Fire_Smoke_Detection/code/metric.py
@@ -0,0 +1,115 @@
+# coding:utf-8
+
+import os
+os.environ['CUDA_VISIBLE_DEVICES']='0'
+
+import numpy as np
+import shutil
+import paddlex as pdx
+
+def cal_image_level(model, dataset_dir):
+ file_list = os.path.join(dataset_dir, 'test_list_2.txt')
+ threshold = 0.4
+ # threshold = 0.5
+ matrix = [[0,0],[0,0]]
+ fire_to_no = []
+ no_to_fire = []
+
+ # 观察结果错误的图片
+ fire_to_no_path = 'metric_fire_to_no'
+ no_to_fire_path = 'metric_no_to_fire'
+ if not os.path.exists(fire_to_no_path):
+ os.makedirs(fire_to_no_path)
+ else:
+ shutil.rmtree(fire_to_no_path)
+ os.makedirs(fire_to_no_path)
+
+ if not os.path.exists(no_to_fire_path):
+ os.makedirs(no_to_fire_path)
+ else:
+ shutil.rmtree(no_to_fire_path)
+ os.makedirs(no_to_fire_path)
+
+ with open(file_list, 'r') as fr:
+ while True:
+ line = fr.readline()
+ if not line:
+ break
+ img_file, label = line.strip().split()[:2]
+ img_file = os.path.join(dataset_dir, img_file)
+ label = int(label)
+ res = model.predict(img_file)
+ keep_results = []
+ areas = []
+ for dt in res:
+ cname, bbox, score = dt['category'], dt['bbox'], dt['score']
+ if score < threshold:
+ continue
+ keep_results.append(dt)
+ areas.append(bbox[2] * bbox[3])
+ areas = np.asarray(areas)
+ sorted_idxs = np.argsort(-areas).tolist()
+ keep_results = [keep_results[k]
+ for k in sorted_idxs] if keep_results else []
+
+ if len(keep_results)>0:
+ predict_label = 1
+ else:
+ predict_label = 0
+
+ if label == 1:
+ if label == predict_label:
+ matrix[0][0]+=1
+ else:
+ matrix[1][0]+=1
+ fire_to_no.append(img_file)
+ name = os.path.basename(img_file)
+ shutil.copyfile(img_file, os.path.join(fire_to_no_path, name))
+ else:
+ if label == predict_label:
+ matrix[1][1]+=1
+ else:
+ matrix[0][1]+=1
+ no_to_fire.append(img_file)
+ # 绘制结果
+ pdx.det.visualize(img_file, keep_results, threshold=threshold, save_dir=no_to_fire_path)
+
+ recall = matrix[0][0] / (matrix[0][0]+matrix[1][0])
+ error = matrix[0][1] / (matrix[0][1]+matrix[1][1])
+ print('===matrix:',matrix)
+ print('===recall:',recall)
+ print('===error:',error)
+ print('===烟火图被判定为无烟火的图片包含：',len(fire_to_no))
+ print('===无烟火图被判定为烟火的图片包含：',len(no_to_fire))
+ return recall, error
+
+def select_best(dataset_dir, model_dirs):
+ max_recall = 0
+ min_error = 100
+ best_recall = [0,0]
+ best_error = [0,0]
+ for model_dir in sorted(os.listdir(model_dirs)):
+ if 'epoch' in model_dir or 'best_model' in model_dir:
+ model_dir = os.path.join(model_dirs, model_dir)
+ model = pdx.load_model(model_dir)
+ recall, error = cal_image_level(model, dataset_dir)
+ if recall>max_recall:
+ best_recall = [model_dir, recall, error]+best_recall
+ max_recall = recall
+ if error < min_error:
+ best_error = [model_dir, recall, error]+best_error
+ min_error = error
+ else:
+ continue
+ print('==best recall:',best_recall[:-2])
+ print('====best error:',best_error[:-2])
+ print('====final best:',best_recall[0],best_error[0])
+
+if __name__ == '__main__':
+ dataset_dir = 'eval_imgs'
+ model_dirs = 'output/ppyolov2_r50vd_dcn/'
+ select_best(dataset_dir, model_dirs)
+
+ # # model_dir = 'output/ppyolov2_r50vd_dcn/best_model/'
+ # model = pdx.load_model(model_dir)
+ # recall, error = cal_image_level(model, dataset_dir)
\ No newline at end of file
diff --git a/examples/Fire_Smoke_Detection/code/predict.py b/examples/Fire_Smoke_Detection/code/predict.py
new file mode 100644
index 000000000..293e57f74
--- /dev/null
+++ b/examples/Fire_Smoke_Detection/code/predict.py
@@ -0,0 +1,21 @@
+# coding:utf-8
+# 环境变量配置，用于控制是否使用GPU
+# 说明文档：https://paddlex.readthedocs.io/zh_CN/develop/appendix/parameters.html#gpu
+import os
+os.environ['CUDA_VISIBLE_DEVICES'] = '0'
+
+import os.path as osp
+import paddlex as pdx
+
+img_file = 'images/000001.jpg'
+model_dir = 'output/ppyolov2_r50vd_dcn/best_model/'
+save_dir = './visualize/predict'
+# 设置置信度阈值
+score_threshold = 0.4
+
+if not os.path.exists(save_dir):
+ os.makedirs(save_dir)
+
+model = pdx.load_model(model_dir)
+res = model.predict(img_file)
+pdx.det.visualize(img_file, res, threshold=score_threshold, save_dir=save_dir)
diff --git a/examples/Fire_Smoke_Detection/code/train.py b/examples/Fire_Smoke_Detection/code/train.py
new file mode 100644
index 000000000..e1b00efc1
--- /dev/null
+++ b/examples/Fire_Smoke_Detection/code/train.py
@@ -0,0 +1,55 @@
+# coding:utf-8
+import os
+# 选择使用0号卡
+os.environ['CUDA_VISIBLE_DEVICES'] = '0'
+
+import paddlex as pdx
+from paddlex import transforms as T
+
+
+# 定义训练和验证时的transforms
+train_transforms = T.Compose([
+ T.BatchRandomResize(
+ target_sizes=[
+ 320, 352, 384, 416, 448, 480, 512, 544, 576, 608, 640, 672, 704,
+ 736, 768
+ ],
+ interp='RANDOM'), T.Normalize(
+ mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+ ])
+
+eval_transforms = T.Compose([
+ T.Resize(
+ target_size=640, interp='CUBIC'), T.Normalize(
+ mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+])
+
+# 定义训练和验证所用的数据集
+train_dataset = pdx.datasets.VOCDetection(
+ data_dir='dataset',
+ file_list='dataset/train_list.txt',
+ label_list='dataset/labels.txt',
+ transforms=train_transforms,
+ shuffle=True)
+eval_dataset = pdx.datasets.VOCDetection(
+ data_dir='dataset',
+ file_list='dataset/val_list.txt',
+ label_list='dataset/labels.txt',
+ transforms=eval_transforms,
+ shuffle=False)
+
+# 初始化模型，并进行训练
+# API说明: https://paddlex.readthedocs.io/zh_CN/develop/apis/models/detection.html#paddlex-det-yolov3
+num_classes = len(train_dataset.labels)
+model = pdx.det.PPYOLOv2(num_classes=num_classes, backbone='ResNet50_vd_dcn')
+model.train(
+ num_epochs=270,
+ train_dataset=train_dataset,
+ train_batch_size=8,
+ eval_dataset=eval_dataset,
+ learning_rate=0.005 / 12,
+ warmup_steps=1000,
+ warmup_start_lr=0.0,
+ lr_decay_epochs=[105, 135, 150],
+ save_interval_epochs=5,
+ save_dir='output/ppyolov2_r50vd_dcn')
\ No newline at end of file
diff --git a/examples/Fire_Smoke_Detection/docs/images/dataset.png b/examples/Fire_Smoke_Detection/docs/images/dataset.png
new file mode 100644
index 0000000000000000000000000000000000000000..65939e77b3919d942371934a3237266b6d58af1f
GIT binary patch
literal 715345
zcmX_mV{k4^(Cm}s#I}86n2vov-@Xfc5A}rWJF=1
zu%G|{0Iax}kOBYzatQzcRY3s%=b_BYKLh}v7R&_&<-`RA3FI7XP0X!~0RXY^+_YP(;MKe4?;`O38R-A!;?X
zKiDs@4Pj-M2K^xuK7GVjH(a;fk2W(#KJD#}x7i$z$6p|T;ZPEkl71*a5j3N){*O*m
z+8R5S4MZOw2!bEpoN@iIfQXqHJ)f1yPiJ>8NZ$B+$+1z5@9$4<22v=M$$#O*S!0Ne
zi-(
z(dROu@(y#0^Buu74)54AP3gb?8DA-(fyvqb+k>x|LY9chFsM~;`vw!45;|@{65fT$
zG=((1l3XM*@x@K6n8F*2j`(j-{i`gPiCP90J|p@*axw
z)zQVe>95{POmy7dt$tRF_t(|%m8Xt+0LwK
zsfi6B&D}h&ynDG>dW0xHF!k0F23%1hoK~vi=YsaB%N8AlsOM5RaFL4uoLU-43aZTU
zJsw~|!uasWiF=_q`v7Ucl0{6Sdr;i*v>^NtVSsu)3Ox`!F>sJSa0>CGc;E?0^}5)jXavJZ3;NpKabOJrwESL>;RY
z%xvH}JqTO?@g9jCV;0b9Kj@uF8%|9~?7;LL#~YRxcFF+V9s>$ryj%tOza)}~ml*<$UHG>Y0WU
z(+BVesR!U+#7{`RUK2eU209Efa-jH#ks%a=aT=Bs@(Ixis|my-UgU6vfpneB8o-L0
z6|@t$6S5O~GwizWMIXH0kR3+`#MIw$FnCX`O-hT<2K|QohWZA~2I3V_cHr{B`rhY_
zy$g^VOE-w$-+E}f7kYpA2Q&b&KNvw^55WOa69OF4I_NTpaDZJt-b}!WSOh^a>@w_l
z2<$-kK=42lDZEJ5fs}^mUffdrxahEGvIuB_tO02~>qzANP$)Hg8Y;smID8{%t6ILiX&zVt6ieqq#fI#_6YJp%D4u#q*>t7Po;GwttFBIlo=H(Fhzu_R68&F!u%ZcO#JEP^7$mCfEe6diavROtk-;(h6zW7}
z_PX`2CL?tt^(75>l~Q#;^+UBrmCiI%7CM{lW><~-ZWV6TVimjwZ=HMtWYc)7d850b
zrOjkZbyK&oX#HcOrs1l^>&DJNj%DxYPYuYjzW*Rts7z>Wj7$vCDA*_>nPBOkQp!@F
z4#s~;J>fp3Jg()TIpC9p8;9Yg{$!(#BaUhMBVzhE#$68$FG%hz?%sDU_d3Vh`+*PluM`jG
z_sS2^Pt|wcSEaY9SE{#c7q@%81HHrmWCWBtn!McdX(Oh~v_gLw|4LxGKky6F;2_uL0y%o#kkUt2$>di7DNoGt43I5v7$!AR8MK2
zbUnQ@?$dvNl;j}NX0wHKW0#<*AimIO)OvJ1PRtFjgLzA`o2lFHn~fCLz;-NXA$GC9
z6y9{!G~(ptA@D)rpFp5&r{6yFkag@W0Ii=dm$9EvPi&}!^0xv}@POoCMao!+z*$m?
zWLF%lVXKj{k;g>TICn~Z3h~J4G~?7}3U}sllB)^0JD|J3v3O^~nF?HmLFHN(wpvDO
zyM07Yx<4bo`Nz_76>+t==96TXv-j`s_Q%>9J>WHmyRzc&EK1(tE3e0)Ycout!C%dR4AR{Yl+&45z6X{
zXk~mCN0;5T!}e1B+LxV!?%db=*X`}w-rn|&wvH>n3*U-=DR?2A3vK~^Mdf@IMa{d1
znw^>kwu)!|OnJ>s*h)Ojvt4(Q&ym^J{Ei{ds0KDQ_6u7DD;hrdD9LEzC1!qB3%Gc^Rg~j5VJ1ezTaSq2{LyXwHR|qqXJ7`o-yB>9dS#jyy+WarT)_VU8=;-{
zP5fQD?#cGgE1iw*)z`+GSYEYHgvYwG_<_QPQH|J2UzX2i`3v%YOnXdo-+#Zx1NmR}
zB6o*+y)h}>B-}MvtGKjYr(V|25qslzP9t;SyAgUYeAVAN4{nufrF6faLQnCx3mFOp
zYlUmccC0)q?_ch?K3%g7^9^HnOMCHu<(}jZ4X?>QN^554b3;cPd-r;h{>!=XLUXn`
z82%IvNhmp#kpN6?)F&
zS=*Wk#^Mbp&bz9vv16QUZhS#N1oSYZePZs1{yzFpVRr%a!8}g#%9xlppEI|$EyPr^
zf~w==$2Xg+!&1`Ei;?B)4b3knzQ4RbwbS3nE$-_*RywS)(~;JXZe3PvovESQ@7IQ%
z>*Q~Dx5K^Ct5=leVyi8iqlXRY-P13}$9UDF-}URAd0!^2)xoKshMn8xZ56#+t`$`-
zUOiphE}6I)YGu;HAv^&PTllRJYt?&po%|IZuH5X3lChqkB^}xEMH0($PyX4;uc{xd
zo7Q1*fZdkqZyyq~>q2bw*vmuzpD&tl%bn)YZ=FZ$pH;7dO|hGYJKnCTdHS0tm&hXo
zHHAJ3l|%f`Jg+U=*iFpz+uh#JA$^ovy4A~X7yB%*rSQef2O#h6?b=oB>TPZLmk$6_
z{JFE*V$s9P;n1nese=}~g!^W+$Vzw;_we%E2Yp5s;>#(>@P7aAd;g7@)g)3@c>eUH
zet+HeEgJWIH>RVT)eT|A97}>&sStIXF5)pu$A8Dx^K@_bKdxWLrEDC!Yqx@ZL>P-G
z7@YD8j4g{@FIvhTMC9R~Ges+0R~GT^%kSL|oigckOxlyr&e-2yprs9*7su$W>tTO_
zCW948!boq`HabBHH04TGrEPtu2HncV{;Uv8Sta7vj}os9>4?DU)(379W70#%kN)kp<#X~`X!ER7
zs#CoTOe8Z2GduM%>kr+sRm|u2yO(xG^^9R8DM6ZKizuwT9!()r?q@2vsREmgi1;=v~C`q8)Dnl9Zd&6BHVa82qZT!TFB#W*(O6|XUkrN{c6tOvd@D3nYK=3>H#
zsgLR~z`b3!aqfalG1%v88HN+I!@l#{(6c*NS+YENTHVEyXj@oDZKuXm;{E0Nl=u5w
z*Z1e$R1W1|Y?K=QY$a|5H6j
zuIza_@IzK#y|Ji&GEfLTQ`jsmN4hnMxRd(+GfX=)E%+lP8AGk;Yq>5zK{}v3W+Kfs
z`gT9Pb!p8eGQtE^J(MLS6(kCc{WIqK_WCA&JKj+}Rv^BB+~z=6!o~34H25y-R!?E)*-o2OFN9HK{1LHd<>odH2xWKOjw;?vjJp9r!HSVN
zS{$-*NvGCOXu
zo*D(sq6n9=9t~)@6$RZAN!OKa?C-hn*KNT4ZtpkU@1-vvC|A-*uLTr_pK0h&QS1z#
z&zo}}+~Rc|Pzi>L9&pZfQ({3!oCqO26^TNguj1RgyKtoPb`5+^$t2ZYdyl?h#aj7Y
zA0o;hY58Si-n9j`8W9mQVO!97eprQw3nlrO;|>jJBj*Xb0xj<}!aXO!ESc;k?aSDy
zZYyNb+*#b$k{kI|}aL~cB~&c){!cg}risw^W{_840CK7KB`*g=lW
z#a-N8pAQeO?^>DD%T?d6)6bXZQr{0Uf0g9FbhVUHQLr%NWlsEq&HpGh@rl4v8uOH{
zFLY~bA1dn}$!jW&w4bPOox3b5GtSNC@hcm3{dLb>pV^*L+m+0&bk+tK0+T0~G#C(;
z$-L1++@8_zIbwMJh#rpL*`C2O$!54OS$vQGzNsR?i2j1*ei!-O2eiLDdX0~&#kAPi
z$Hm&j7B@PK6wgRRqQePcoDWX^sT;~S0OsIW#?wUP>|qb#lo$2S1~sAPpoQ;1>^+1f
zQ{NEcB_Q%zO|0&aq}ISVL#ZOpPAm=u9%({&#=@|l3?j#6bBqp!dwDypANeJc*+tW~ZA-?Xg2@P!|dk_}<73Czd6=B*ch6G1+<2zb79{6AUWb
zE){yve^8nglbzkqxC*D{VwoB1pQL(Ad8F$`^!t+;b~U1nhafPaUy`wcQkKi(l$A=5
zE?>ua&(Essr!Q9w4EjSx42s|WO?qrTgsYT&%z
z^`7LIDTyWNudVyv
zcv7D-*=v-4NfwH-Y!u|pc0i{u)w+rs4OQf3D9sJqyI%_Nys_&ZWSQ)ng4KX-cpWGBIXX|r(I
zmSd_Ot?MZjWLjl<+7>~VQwAj%U)ZGbxs2}#_L3X=@{J(n*tXamhun}k9V}Vm#p_Zt
z@q2Bxhx2mxw9+F4LWH_IgIqQaR-`bPzMIuX!sq9ai++TIZe`nJMzQ9si~Yf8Wq;@8
zq652VRf;74a^k_!g??MxqWD#3<1j^%v~b53qr)_*K
zaze6NZ0+?CX#@qn&evUhq)Nv?#(+*=hjYPZIwxaI%ji=0Ri*i(;Xif
zzR!L{BKo>4euYsEK3__X#2to0e+NwR;^^v6<`B|NMh`WqLNi?YB`6u@o-gg+>5*Nb
zkeckU@x!LQ&mrVvQS#=FVx6ZR@6e{sVi$|}n
zAfnuLiT&q-5EOeK^sznw0}?Zeu81)hP$6hVZg%a#f&U^pzzW>&bu;*08}c~w
z=+w3qMTGS0!IM2p!6Xp7{qBNh?KXFRAJKdL3~IHH4;FA|@7Zh8$DNmFX8EIhBn#ki
z;Y2j}h5G)yrtT{p>f`LnlG)t6{@j0_ba}OyA;1<37sP%P4gq2j9CIxcZHTvl1FvO(
zzEa^Cm1oZzKhQNz6x>DxSBClN18=}p?y;Vta{!3yjz7f{1TTHH;R*KoSv`TUQTyh-hsY+)*?hcY%J*|Y
z{MHs7^fcSToz`vgAjW1~FHm#ElA0B0qVFo&0)-TD)1s7(XF7vN;*ipPQE_Z^9G9NTydh*LUZ76?T^BJSg`gET
zB)Vmom+izULh{{$T-yen3;E~gHAv76(fyF12Uc-Hz3>`!@!<^4^xAJ`t
z!Yy8pMbNf~Toz+)nb5ad(Ph!rkhV3mCo3LaK!)|4)t}Qkg=VWsa&G(q1NkPyr^K#3
zJzO&1-z#l`;5i}#Pt#s#R}DLIzL-pE9Q!~!lHA*B`UKMBxqMPd%AkbwL$(2iJsp7m
z-ITuI7HG4rT)ddm&a4gmX_U>m8wxpc`TlwA>3#a_vS#JW^KI23JHC0C7G=%Bbk^tc409&K4Gh%VCL-5ZSrS8+Thizg!usPzI3VP$+T;l|Vn$w!4$=q)=g`T)
zYIq(9DjuS|y`z(E7v?OhfMjps$1dEj5}~0#erwACnm?Zgp(G
z-R13`$nK@pLsq;=r;3UC(K%SNJ{{oUE}Ya1N(DU1vchr5rA6s=5VPkG|E0-}f?&wwWLRf{e}K*OMZy
z1pKSn{;2;^>S2EbHoW$9i6Y{flvm8+MI#|;5Kf{oiDq;!-DCL|F}-HLUa&@`qNqM7
zQY}|7F^sMKAtD^=wDENDyv0#h{z*yG|MOC_HNnr(MO7k=0ux=M1U}X|rQsh=tiRIg
zOcGn_<8abHM{Zp_(dIy>kiyu}*I4*Nh9GrmyQ9Hf{8n7Xn;uMNiQ!AG)iM?_j>Yxf
z%faoX?P4FhQ9?KVS3nImVPSum{_FMEhX$b5GS0vc_p{fF7uR-k(ASgGR=2mOa!+Rv
z{(Je2k0nlSpe??cLS{j`Lv*d7-0*f*O7SOR_&y3WoZnGzP%|ajlOj{B91R???8<>ZbwgEx
zZ+iEjt^mqaO{v@gfn_ElncusanQVpXH#-0h)ZU0TFQzS<1c`Rkbd
z5M%I`HopS(mvupA7B`(RiZ?6b{(WxF7fW~_{Mo3T%3C_V02WxL
zQ=fCbGJi3^7|gMzGt9}F4oBIj5B7Gz)DMp{!&)d(b3LdbbekW11BN6tDg=d}SuWqx
zjd16ESRM;B7*4JKTR&iL4f|I%c;X!frNER*Vt7OB=Rx814BQ&DX`lf*ghAU+#eBeN
zjh$&F9@Si#k4Ram@y@IUB#^ig;E_#ha$~T7~)K42@=!k))wd>7mA?rzRUoro83f3R(6-FH8qH|dPL4Q|
z5r5KseT=ox+(2Hqa;wAzfwO(07CmjkagIh4%*CE#SF=i^X)@(xNCdBQ$GU;Wv`=1R
z{8sVuJRhS;&ATe`xf0U3K#Zj^++wA4>7`6i4x5PtiX6&tT>`&%I*kWdJFyhi#)jQ51aZ1iJWpKKj7q$m`3
zf56v&WrRyT#2?v*>7U&+B`9%H7K0JQvqO*9jE9(na)<(i`b3Gcbp9ACqLXvXTe)4q
zv^-*hp+iNvs-ULL42%W?=+Nw=Fs*R|mMW|`ooPDwRA&5iK%HuD5U-hG^10YAT}Dv7
z5whTPp(&sr_ob~tQ`4s--jCio#oF*oA0yPKOwKGKg3gUp<%RZobM>njLwj(BVkV?`$u~iJf}$Vv%YBMq)kq{H)Ddr5
zB+ZqQiQ5Yut-th6h@>IDO?+8jfmVcX@1ebz*)zrdVpPDNa=wxa!tA*8B@i2j10}xU
zbZ;qUj(n0*<~Wd%Fv{8Y(DvACh5MlWOa{#BQWIbMg!8=K6g@)rO3$HFnO$>d;KKiY
zry9YD40%*58NrB&?*rh`BL3v3cAlI`ZFf+&Fu#VUEZ<{o!SZTK+9wQrgKI@`liw?f
zh%xMc&DW3p&~YNaD-a%W}+MhUxbXK?+b1b63kY=
zTLzP?fGQw3re@I+`7g1mfFMho+HFwI(b%+O-`3p7OMT%;EDmgP!`k^`(oL`XOB!ZW
zQnj~ejP?C4>hJZn4qPT$+lh$d`dK4R!M7;ARI>$}bjiGdn0Vrc{5|+7Gbar*dG2MT
zR_=YRQg8*9_^>htZG{RMr|bKPfwk#1Go2^6PPMi(=s-s_e#f_}3da0dC8eLEW;#^W#umZ8dPj
zH(*#3I>r%u#lOdzC;;iY^%&7&eX)BCr)tMm>!A+AiQrZG52=qo>}`8N}1%6;Q9a($r0`3VUig_pZI9GO#J6P6UU%fLp&hX
z!=*a@${h8NCGnAuFSjmlBQrJ*e?Ed?{)&LyOW@nN;%;px^|Q;uwim5@o|hH)yd
zr;^|gZA21}m~qkioOGyQ-HDuWW5MQ=+4&g71PTAQ&wkHYO!9{ACuc;<>=j6Z28^0Y7
zq{3~waYd!6gD+yIZ})v7B2NGZ-gLbeMcKo(5(bF1Ff?P(`cQ+@7KuU~gXD6vuwBoi5%*?(Xqj6?knDLM;XiHGu
zh`Wq@n#hi}^?>APY!(ZUt6Qzj6sJVuyAIH}n~22aGw
zO-GGXGesmzgOv>;=yYzxLNCdrJmAgAcB6mgK^S;ed)oJ54K{WLZ7l9
z=fv0(t|q;P+$ZcUs7KIaWrSB=6G5Lvtb_C$$Gn3!&U2DI>)|FrzLUJqMuM2~(<>Jr
zLY#=n(ii5Osr*>q_7P?mV_)_h3(MYli&>HeRBT5?K$an4@$1TkK{pj?zL;}OO@HB<
ziiC?6YA`F8y?{mig%KHG4Dw`gJrR<+T@3v228g9sa;AcXI?LY5Q*#LX^m=Cs5Lj60
zzf@(;�y*FXSxfEzfn-8GqM^I7pPg?K#saai9C73MQoJrHI9pW>@SCC@ts^>&p3f
zX?fo@Nw^+Xr!|cYr!`z0$F87R)iu!O5q27sWmZez!^-uf7uVq9lfS4xfOIrT898+I
z3z8;}kU0z?;0o|d6ruo(RwR;RTv>L2;$x93ah7Pi>Nk{s$Sn`?8M6V*6&%zSeQRv_
zae#n7!1pX9$)vN6+JC+Z6hNYSzGG0vX-n3Hz5QZ#C&>S3CLvb;_9KmCmRd=usev2R
z>uFSot%~-F&Hk8y5(*<^lg?46l&TSpyGcz^grsF)Di~XeJa!O6VA|KN(@wRqzX;9;
z4|Cl3@?z=}2IE#lcHi{Qi-JVmQHTJsHpsBo?Ant&X%5!;P@}$6sMT|+R@`ebF?)_cQEzl@zmD$=k0ABhud{5U=6{2KqsSSla^V$U4&&@lZs?K3b>PS
zLSUy)Z$#Y0gT*{08827Sz?XpXH0}
zssM^t01`KI^m3+VmR0{eHPM6&HsI}973f2@UdZx~^T*c^3WyumfugU#^0tm16d;v0
zy>DMbB#n8q`t&}fJyq!uSvxUG$nf6}U0~H@3%mpkdV11(rFq4}?It63Yh6w|d$3Lh
zU7AQ@9bVSBOs`g@U|04AzL(u-Ot`L#Pd|j`Voj2R-;S#ApXBLav*I5qxoiaL5Ry0~lSl9cib!J#fCV3I@*U%(&pzPW;
z8fTSwM!@bILVZc&W@JFic-n`##LiUQmasm&jN=%Qi`G)4F%mm|qr}W`On%up>ub`K
zSKhRx?J2U2?(Xguiz9f+DK`>beDoqBALaQ3DWs$I3%@7YO1D)>&Hy#Zycfz~?{z5w!fjRtU&f6(p
z01UrIu*drw#QZ~gYT%mroP;!9;i{CSl@nF2@B3me-tN!&_1p9Jb!mudgbB3pA3M%@
z-H5ZU*!2eda-)rLb3Y0Pj>f0bux=tRgvJsWVTx0RW#Qx1@EMU`clR)=ytOU98Y|wu
zu^gXb{BwQNdGPzJaJg#gkM#Nzq4WVymY)v$Cy%nLL;z)S?bqp3sT&ysb|BO#olIH_OF^v_iv{F*c@mUJsH|K^e0Kl0gzh~|wa5B-le#4_lSFfOMs@Of0q9bW@6zVKgN_iNLJyJ-+EAd|eU@=TbuM8(rNJs<|u19mj{z
zANVU6im`&(StiBK!sRy+YT>;3uzr~?)b`Lv_02*Sr!fZneWtJy$_JQcal}LSC$1$B
zhKa=mvAJcbvhz4|Dribjia6?OwBu2*;d?}+*5jSS(0$&4Hs*#Amo{!V>`ORQ2Pxv`
z*J&-X?~>QECZtx%dd>lunJO~ap=LGI#zAFB&P=t6n+IyUiMP`R1J1esuDGQ^c%s2X
zb#>rDjA&K5cVJ7af|K7Bl}F_>7TV^DbF|CYs<(Nfs(+#IYlmKP4(6NFz9Dk;~+Y
z#&MqtNmMh;(Cp+K=cCOV^a!iV0YPP-;ZtX>%zpHEqcI~$3-c0$27an$B;cAZ3E3s8
z|K(jq826;v`dJD8xsUz*xWB)i_5D#aA9-J7ax(NkKk{k`Ag9;7O5$uH*k+7X+2K9U
zGv1>{VNC$EU4;CzXk_5hjmaU-{K*-$Dbo>^u7|ak+QY=)3+VxcS~Go$yQ(L~+lENP
z@5lV^Krv~jsGS@zEQ2icPw`%$7d%T=-!Gqwqv{Dpw(fVb`n#toYo@hFJW`hr+?)wl
zl^lZ1B2+CLPd*qMn;=xOOoKA}ARj@xAor$`ZK=iGAqG!LFn1B4FUS=foh8yir`#Zs
z4OwR0fOZWQNONL)#bm+qXfCQ^DwYQc9f*677Db*Ib8JRk<3RuvS)%vK{aEK|OkeC{
z8vZ3XVDv3?>BvabKg8TfvG9FkPyd)}5QppKRad(Y`#Ef2u*&6ncSx*lyiYB5D)*v(
z_m;h>AkA^M*6o~Ql}rfmTQu*{)+_1r%Y@dR`fCRK8<-BD{QXvy@x}1WpVJ9bI1lEJ66*vC#*jsh;L=*whM@C}$aU>M<{i@sJ
z+9)i2ukTu?m0Q*4WEf~z3Oqsb-9wf*W_|>S9z8vG^SZtNxYm^t`!Rp*`8~hfku1P4
zZZ~KbRbfh@&z}VvXj)lcc;&Bh3+RL)bX7&;(VjvrP%+cP@6Ffc
zjEh}D2p^5ZNiaNvS8NomtG78?q+?SwCtBIV56N(*5qSt1_ANnnY?75`iv80V(LkOh
z+y18fcM{lOF<)DE3l)@wmJIZO9X$lMrmJp1^NJAXO>_9bni;g>R@`q~h4
zi&-q_9wDdv-y&iuuuKsFU3><{^=e5NxLnXJgMxvg3a#bk#?u*cvwfDom?GAloS}t0
zjIGBR%UK2UWLH&`1Z1^=x^Gx!$0w=mW2$*CHoxe(I?su
zJL?(6m4#r#^rR}KJ>~?7v0EN(eKHD5sj*-P5}(l@L?fhVI&Wz+1{m4!jPUPq{CQ&P
z8;Jz8(%V1q-!fowsd38&(JpZK>R2<)x!S|E2RvF@e*#GC6qcVB)M>)w(&QvrmwnZ=
z6HgK^>=Ur6!1)Owm0f7ZLeW*hOV)CjB!=Qb8WoKh9Zl!rH&JXI%y{39AtHojpGh47
zDK{cN!@bC~v<%?CjDhi|E!PCv@+S>w9Smbbe
z8a{^93<%Lr#*FN1bYGK%I40HN9aAm)_N~PmCLU93;sL1Sf~5?DV;nupNmb&w^8c(u
zjNhK#fSU+%^jn8feUfJ;feEp`E!~g}5~HV`s`n6lE$YA;w-Ym>kfEhfr_oU{o<&WE
zL#Gs(n7G61KE8|lt=QRyBf*prSh3?E>?6jy0GP)pnp@a;>pDY+&gm+`4t!KP9f$iZ
z;4PhJ%=!?}ZA(AJ{lVcQ{Z-|9R&*H#>O_5s2;s8ZTmu%j9>
z!MP(aF9||;8xoa7t2{7zsX_#r*Wj7}N~Z!VcAx
zQUBdV#3#g7ySikrED)$2%Zp>uN&04VvfhpT9y34Soi;M}39NXe$Zs7#ZXIu-(T<6P
zj+)dy378?20YsPV^fJ%4;t@BiB4{!np8JeBx-KtvZ1A|O;u|f&1{yr7j)bwQTl*k%
zL=odamHmPh&@#47wJH;5Ji--Hg25A5SE8bKo9jWu#P&oc*Y%B#^pFE1;NKq`)|Zll
zRRf~e#Y5T4Ghqs2z7b5If?DPT&V&`N<2nAMCw&+~&Xe!VI4jWZii-JSOKM9J5^ETt?VdC%l*
z$CRSjqAba|SS)VYY
zDNm+HqQAKzeL0*!(5#+=0zkp^tMDOc!<%5}^^<)uI8}n2KwmoJxw^J}|I04wxfenGuwlKnOUzHq?hUS^j
zTb-}7Dut|!ezZ&1(aR%gE)J{bB+VA<+GO4+l-d^t)a0x0#xaOmKe)ewP6ObYn+9ZJl~|p5E}}1`cQ$=B6`h|5&B~Chb<~5?DkakG1DeS0osp?Js}pX
z>RL2kC}LR|7G}+=X*oSnUiAovkN#1=_C@<+tSNrTq+}UG#(_=d^gNWkdtp3q95+L+
zdDp-qIYq?W&z9ljkv}MD9o}!qr=W1lVj30qRcOEb%(i~VV;ZC#U>=YBi%QAZAH>j?
zqvj*qP7+3>e^6O4H(+1-V^?pR>D++j;h}4^v?o&7#y}Lc258XsA?g
zO*BL9;wUZdqtxQXK;e7<){0T**pfu4+xs2t)!=V>x=199Y^NM
zGXlYq%#nrJ8$hsFkWUCM=v>8|r%$30b=2dGd
z@c-HjFB3+Y1KV1aw?n>3m1+dCXk!J#U*?zm%|{}zF7~(ELqPh6t(>}7%=ZfD)(KoX2p}eY*uJBgc;4&PJh4@&ct_RdBXex<2QDV$IkUT73D`
zq5(<6Vr$Re$bqgM12M{iy-5IqC@XPK_chm@;qI@+m_9VPrGtu&=H{Hi?h?V`j-arn$#%7cee5dxX-z
zxA^_3%Cq31$LGmUbPfAJ@r0R=!O~PK*7w7ntccY~QyWE2{k~ofOmVG`6==4|kAg5f
zy?;OZrMXQvB@3soI?UoDSG@5xH5)+0=)g6KB!W*wD~4hFp3d<8H>tC?A+R>+-P6@f
zmL^Us(@m61PmA8C<{zuksJ^?drK^XZ=c%>NCwK8;wQR$d9Y{k8YljEJ0g=5PAax
z7xM21nR4rt4BvFaSthFsD@7KDi+k&c&zf?uWpc0M-audrF6`JfjH^gNC^~;2p7Eay
zl^+vRhQ#gZs7GZ0>`cV(Wg0o?@ANXL2U^0kpaS
zzg;-fZg2UU;=71IDa$lIU`k~q1s3z8Mtw2M7pro1ENh?gL<)(ueYDZBNv_IWY{_+_
z;Lj8p%B@uhJ4-CLvzEL+&g*lZ{r1<44FU+i8Z)aBD;R#a=giH>$|q9vvxdcJZk9JU
zHz6sbw~uEkQt8HTFD}7KFS59PHMFhh_=S?GZ4Phu+hDTT;lG%Xl`2!vN>uLAr+S0s
z6^~`Q0?}#iG@#RgoQ8G-5&P<~&3cT)o?TsduR3(N`!H2cMcq-^HWZta>NxuI-lJmC~y8e}6i
zC~~q$Athe{a0BGlG|sP1=JeoeLa&4w)ZKKoepWeR>->V5(nN9f#_ElfJl`)9QeZ60q#i&1gj-C{an7>J;p0B#Kg@irz>IrfLa@;6wri#%e4N%q^nu-HP
zC!)ZmJ!p;uRm-AMst`0Zd3iAsaNevbAqsu={+*jfGzvf6n`$pcoa!k*-FiznPE2l>
z2N#SEicM
zuNl>2U!;Jlji;?QqkwJa9uF$TQg=m5wB)qH@tY~y)ILRs57N=uo*_u}B*zA9S5p`xrxV6iV!eq|5#>JNbhP$v*i*09Il0hanX-RgHS|
z<|I|P)iTpU;1(CXNR~9snRvx^H^tI|I{<4E(3+R^#7Ot
z(Q*98>f%4RnK?3ZQuN^D!|CvEucGn(TmJFU;JNi0S#G7eK4QnSqx{rAo-hoR|A0}$
zeALOVh+_(?goZ}=XFM!gOu!{XoRr%+{iLtn5>fn~u3iOYqCaWXA~5(TVuxB5m%FZAFM^4Sk1|j%>lMNHl6G3tNRh#THoM4^AVL%+q{wR&>}0
zM;%0s`9q>`n2XwPjx9;F(ELfiHRF?rWD?p+vQ%s%Lz(NsM?oCN05VQFR3x)Pn4BRJ
zpGYEDdNm0aU%}4|NN`p4AszL^W*5N~ks`KmUmJ-#on+>!c7_Y(XYIv96Z2|GlCUoL
z;-P_hM8G@Mf^a14sQne#rjx~1{p^HV2BF+<>9fbWeA3`AzA>eAW;(PHjIF5`qDf4S
zxYf6CsTSrk`>WgaLZBAqspV%PD9q9~QdoL=Ct_PN>Gi&9MmC~F>8kDu(wqVUz*T1(=~cf
zHdAHGRU2s>{@5~$jDb3onqD@OZ?T2GUtEshL=F0$oWlQl*gSAp8B`UbxbERPi*d?G
zb>PBRI;g!&>4&ak*98h2o+f$d8WzGxr7UG4X4Vn^rtr}ivu6IIR|hCpxFiqENsIlU
z!)GjFfUy!TG_~09`0l~7pYC5iKlO#*P6yAr%?l!QLgGUyLVf>8ZG0t%QG+#rqyR>U
zLPLNh5Bg^0^!52*H?Wv_nSHs*sH(gVa9^dqOcH6*C}qE2)4rhtpPyPYJ^WrVB3kjOW%0PY&9C
zhOY(ouC)S$-q386b1AvY!7z&BelQvRc0)o+#8{5jn2jx{s?_K^91@2>rrlezpjoe`
zw|P-;ITyJ{QXf>A(W#x)7A(?&>hvu&6q2*KLg&QK#8UQmZn!5d{D$R*ui?aWnt_`h
zQ!~IFtTElyL5>R&e@tm^hCdaW#CFBMRLB~`FBXxRTsm;EULjQNWu{5scr-d@x=y)*
zE%6y<%ju?uUCSdXhC
ztQv7ULDv~ab-7O1GRgd{hpBwSX`PsB29Wry&v!$}n`hAWeZF#gBTVKojCkkzl7d_TJ;lA#vL4-m5-oqfta-|V~`-gmg~i3Izr*wDN1S68=4|#~wfUWn;a*
zaWzbPn}f%ENGn>>mKPu&5ATot+xw`$u)&pgzIMRF?tGcrwG=D%p80qkd@ihoY$E$`
zJxlCUX+p$SpU+En$mU#vB~*Ci#48le@dVtiwY+jUQt_(b6G}|9O2Z1xAhZ~oYC%<6
z)7mX&8!v$*6t-Z@gy3|gTJ}q#ls8OHmK)VcN4kw^M~C4K45MJhXiTfgmN>|hyz)i%
z-bhu(Q$nUU%H9(*j`3GxMYfRZBsKLx-FznoljW(>cIRMfKQAj;>pIf@WsUYzeAo2~
zrS$mm-rt}uJQpoAXg@iy5i1y8&3ENVpB!n!wl*nrTxw+mi-8qrI&;A0GBDv#RZyA(oqKWaESU*!jF33
zmBOb(f{&}|vNR+5en}l#jWQY}xDOr~wT9Q3Nz8)k)aor_qI#gK0Dzee|Ho;nPVh37
z8gYq=t5Js@^8{+jpf^mDnX+dzlEcpBt|tZF=YA)?vout&(Zgi<&)EI25T@Eo4@6`l
z?UoJ|M3|=b8uHW=51^f4DFtCIh0oJ94=u@^B7McANC@4z7i|a|a|n~AzQ0*9%^@FZ
zPbamCF<1FJ8yspBhf9rsZwSsldsJHAVGwD}<`-Q^JsD=Xnv=jtYOtge=bFA0tBVQP
z3j4L{ro2DR=tAm4MfYXhzW-coMWN>Bde_hj#5CIv-g0I-0}9aU;<8Ju>5LVsfQuSW
zf_kgxDTxQpxM=(u$VHgx6LO5VIhf`ETpI5IZM%bD)=_^y#!!4DYqZ#W=Iom^iw7XX
zkzj?b;{CF!QZTl}Fn;AfjWjHTRkq(#m0kviYBZM-WL_Ck(b;)^;M@
zKmOw%xd|-ep<+gZdpS?(12GRU*9TbQ&p!JbM6wpgL
zRv(^`x!(JQ=m>M>+nNZ5>w_+#?D`Qzq%_
z4z%CAXK>Hj5s6C4lWFysb^~Rw;jlPyFRe5Hrge%&IRqz_5?&P=I4qFpI7ot=thrKQ
zXZS5;$oD(_;t@uz6lscW!olEKh#}DL-2&40sYz&|UM3b6weEPZ8aI7f3gPA=1GW_g
zI4rHy_0A_G3O}5I;kC=qNl#g7^Sm)GV!;Nc45V?bT1h@UbjJlk^wfQcNgHZ6
z_{YbzU;lzj7!2F<&|4)7O@+wtN{*sVRWm22^}|_j!6vU?rzNrG!PAv%VsQ`hsD)`1
z=*VC*zG~e7K{0LUuf6_%td+mK&cy$$L7#h`VVk)Lu|oQUTwng>QCRlZtUK37p0n0q
zL8iE?3jaP|el`w(Dml8@7UHOx#8BDF7K3f`$2yMnMI;h&vDYhN2-6#!SeB=!<>_J^
zKwl599*l=8YLFyI{TW@G;|Kz^v0+iKr{~Eh^3am?y=EvV$I4RgZ1DOdcl^?%VW)x7
z+$sqr!b6?B{$CosjLR($yHuOSk;A_AaNz_
zW_$^o`(!wIBUsX1O6A@(ot#Hyrx>IDh-)arvVg3k17dzu$e6
zZ4nMbZxYbQ@olqvhl2MP9De#~o&c8Rf8g-;-82o5jO0u6=qfc}Z^E7hL(T>{_AoZO
ztH&&i{K~&}cvtf$l>>~p3jhE>07*naRQ(PAfWZF{_$&V-sCS1VSuAPG*d6vUG9vUm
z$b^;C`19QKNBNTYk`7`lWdNieAs#fv0VR!tGC4;5Ds+(r8U~K~I&31@P@DuVuvK=?
z7&87j6>qT@tIbvAik*7uM?;Zw24X9!GK~o>IExv~xfuP0!b*bkGITP^^edibqE#6H
zp7b58;|dGj2@7_$oQIFTSXk9WAu8aCGesUB{~7%#{-7T|g@n-!yB94LKcR%KoS|p@I-_2!%&_TzN`F>Szm(>0CL1275!L
znq&T%8#^0+L4B?7^P_w`d#;WMXJ)9oiqMm=p}
zBUXlfwJPsJ{#7&i$i!ouCAa;-fAC%)DKq9%znTrI?XS+$5!v!ukrb>!Fw_-MW~7+H
z7?2wS6M^XW*xc|!>?+0{8dEv_rLA(GC}6Y`wvltOJ~$S9Y|d9WvsJJ8e%N7+GS+TO
z-xX6jIhZ{Tnh%Y*n%D+w+k_lep5lCIa#w
zaVk#S9=3d1=&g4|DEm?=g)`t`QL`HAuf;dm>TExqNjDkpq-%?zqPK>f8`JZ@bkYZb
z(hE}gR|G|DN%qJNj}n3kM7H)c4`o3?wUPiGizh*S-J7oC_||6?kD3?|8Gie%hC6
zXq|Q@J~`DvQK&dML~AHVpf20QsvKvg1hHI~y7Jd19_prjwBp}ZkkjPj^RtA&@(#@A
zSIht@Y|)68!%spIY@V&}uTNU0W^iJY>2fH<2V>
ziOil|Q6Of2-)7F03#G0Ze#2$j%x0k|*pml7si#R6rP=0
zNj=>n!whpkFA^;@4ieO5e5)S>UFez!ov=itILJ%=)9wU7%~Mip8XGT+u~jW%vim<0
z%i@_MlOA;p&Em|3Vyo=m#emGX7{f#(I`NIVL!UJ9KivV)CB24>#(#9Xw_ANKVq1?f?yD+DEr#K0|_ZJKLtIJpw}fG4R_hUuXiI094?`Txk!pojO7-CC)9dpWbe1_xeqIL)|}t=x^MYSa}|`^
z6sUnMmb3Cg!e`@6UsQa%`RXXneDQ>jg#Ye6`Pls+EkgU8yN%vfY0zyzTySxM2^C@ZE1OR#@r)jit8U6m)#|t+Q;UM@&mGiXNP^{;@IX
z;y+g7KXKSnxt*uN$>%4ztS;+;cIr=)mp&N8@qs!&YiEDGfy_BC9epZwx{
z&dj6YW~6B>XE*+SC}%`$?aAtQ;w$s=x%h8TJM>3VCr8$yUtAc@va=fOcs>WDK~nP?
zj3|x9#-ty^ZTrJUK_bB$j6|Q#Dq7(u{h)JdA?x%U^O(+5k}z&+w`#S8nn*#a`=PNw
zfBY?koL=}$ZX@EYnbt*m+*FYqcx->@YNKaKAm1You~gU>$|`79B_p3H&Ea8&YS^61
zBdkWJ%#`Q5a-yOlt>7|`A!7eWV+!H3UVLyUZWR231PcG5k2FF;Vf?IQDb0mw5w--W2!n;VNCOa?
zB(>@?25z_{&^X?v3-=E*LpiZib&LsJ=|xxPq%ux*c9}#hsp11+F={Bpl1&vRKWb*>
zAX9iGPcYQ5#{0~R^xA>?Y3~`}$MsX-C0feWIVhzy`?Wk&XLq>y+aHa#a+MUCd6ZZA
zrg3E4`uD&Uf~|PTb0aRSV>?tmbUgx6z5AzTQ*?-fw402-8OFVAkBBD*dVN6o>z#B7+Eo?JxDDyP+mqtHRb6{##x>8?hy04L))
zRpXxx(0j!)Ku(hn)k{D5&{E?OcV_UhOoU2dUDns?@@%=8)Wl?1wSs>5$LzA5sL!8v
z_!!Q-gJM#|Osdd1gSBXsn8$wrkR*zx-1T@))us33BRTaBk^SH=W?4<&J+u9tKFD)J^7=RuCMSpo@XZxqfg$xLwKdVa)UqK83_KmX;=$Jh7$`%eg%mlBrlmegO*uRWKG
ziV=wHioDevaW)0B9=KRPxd7?UF+JKd2D#Rw#UV4dIh?bip0M;n4S*n_V
zYvapq)^m3`N?y~5#X`ecM0)kUla0+8nL?eG7#}H(Pq$ZtPNv)ZX)lBJ7jUqN+4TcO+r
z=n?Bc0IVaLufC)Cy5KO77cp|P7jFCFpK-$0u-U!2dKB=yyqVhYa1<)K3UdL?WNE}E
z<1TqA#bXU23vdh*R(}cP4i~FhFGiEX)$)2EN?z+38Fyaczt7Ug`u4P64E9@%TGR2=
z{SO{AkWel}us7OvBiU0Rk24|@1E(M*kfA8l;WDFvqzlAnVk0$LY+=lSEGvXo0qt6%
z6Ag`ido-SVFR@~<>3~rHJ@T-u9<@|kNF`>rHxKF&j$w)F$uGdLVqIuHj7Y!*$T0ayLv85!r+_p+jBimGWMvTv9NoI_cS>_1(sHOc}c14GR
z3|X4^;TEJrKjk;2l)o^&zVyVbQc188xten!vY+}0Bttts@txF=UzW%X94Zcndbz5s_#-m7qVG3q
zhla>My)Q?s6!S`VejF8!$tTD~IcVzL-->0QkDvbVhtJRN_p{s3Z)iiqQq)<%}1I?MK_?@#*4
zJ0U#QVUmvx(=Hkciy^7q?AN(>(|4yHy(MF+L+FW~HeB+QYxc|#l(L>g_y?x?)E1s0
zVb|H#q1tVQkEm#=Nhps)77$u-j(K|0yv{`?G(wL^6E}hYEZjeZay^m_PBpLXq7CR0
zp2WHE)Bf!+5k-k^w!db;*gA-YmgTY-EoZ@0M8%WdwQ6e9>n7{$>C?AtzN#pBqdRIg
zq)Xh14};_O?d`3n)#<@&l&&9MKJI^6zf7ga7F(IF7(NRIZ_?wZ?$SA4i`~gIx1Q0vmpIS>zN#!G~-3lo(t2#`k
z#e699TK1g-sXuIXvX}Z(Ue<3b9OV{cCQL3Hqv%B@9cAy}kD1gtFt;D65SedCb)c+c
z=qZfZ(g=qMGvzcj>^Nh~uB-~kZ|k5dKPwDasn^d^#7}?N|MKmZ9;)Cu7Esr*u$h)o
z6DNNgbz^qos)L5fxS%q@d8Ro`5TM`llwZDn)!}ZAw>MAr*IMW?AOB~@pfnm^q!0?q
z=oyG6ideot#*&o=AFREX%vjgm(lyfbb$iEm-~8Im<;L>+$K`Z+eS7yU6J)jUX4J}H
z&KB=||M+-&dk3a&5#@U4@aawhY|BH$O}5X)Lr{W(cQ1=8X*GDIQ|GyX0-oiS`J0}c
zTcvt)uao!a?b(_gj9LST@8jOr&1O{Ts!$s2ZIgBQkHZAc6}oj&fx4%z1Y<#J`b*`X
zr&Fzyp>M+o>l6diA!xkw{`=+IpWIry-hQU1$K%6$?u7x+koJB(`BO!>@~)`+#_M_t
z)Z5#sSAU{&JrtI%mexq^iLL?nCSZNvyCb66I=T+S+1o8T74`7|K4qVflL9eQZ6vKy+DFnyO~f1k9cVaHH7QV^
z&j#YQE-tN3X#^k^w?hsOD_cTgSX`=Fe84&0bPkBxBd|^+`fON3Lrbriw^n_WILU?+
z_*OzAv70eFr}wVZa8u-og{}rRrneXuIWionget&g8^MU!zDmlXtn;OK-hwTX4?rF_0ecqMuWYpYyWzw3h7UN~PQi
z`t+mQbKH&4F~=p;(qszzppOqD5Q
zrvpuAF3DZ?B*G$IMmR!qLxHTX
zp=p5KfMn+j5f63gr7vX=ljV=c$qr{Qrm%Dep236}&WIi!2LB|78?VJpA;;De8g-0C
zh-rKLixf0ct(XdkRvCv3J!cp>Txq9<^%;XyI0p&n15nyn=U3QrY%OmF?R-5^ZKSl6
zhy^YAQkK569bZ@kW8cq80!LH0tVnT(U=#%&aDKx=<|F
zOkOS${z*$~txQ1=0@^4TkQJ$;_iQ&bcBz?xv5D|i4RON@s3;bjkir9eOGyAfd3n3=
z4>~{Hwol7*z|@!!wJW1+Ur}$58slFX%9jVrn6kRTAJvh*F$E0*8s6GfOUvU_3Xby8
zopxrtgIG|GZ#A*xd3WBO_#t+-y|z4)L$e70Ngt+}!BGamOi#YA^RO%Z
z?KL(%teob5-v(rzWlo&Mip^yqnHCdFP4+#gL(&$LjLlWl-qR5a0K(jdLK<0$FH#A1mfpB!d2rvPXHvC4?@hv?hekY3O0eto^#1os%SO5AWp;#&_~iadyWwPj!CerDlgb7Iftr>|ds`Q`iP=ZD!Pck4*Rl7a&iF!6?W
zfITiENPy&lTT@hFAq+e$G5hpbZMHd=sIC<=zAA3UZn`q)YAxiN9gwYV?jFv0E7T&l
zKfaIFRG0J2qXUr=q4G2$_0>_`ZkvZ&NcQ@s=~+OT#zWYDh5Zv?Cb&B>@pG<@>rzVs
z)pq)Q=ATtgiPcY?Fe?5}K+e2wz6I}>Z|-Jt3uxc2pUoiBA}{iOt+SWtSX~1gHncS_
z-`mnXk90VBF{by2@%H2U+vd{P6#X<^B32(nFpz{raJuT;WTIs{hUqX^>e=Kloko}jNn)<1AXaPL^GO-KZC@D86gx8sEEmh9t19@9
zLq}ssCTIMkHta?+`b7^#lP_+aLclJ2E>cv!S_57y)fRojFzgOrmHBmB1XyzdEx2l>
z-AJw9nnu4fokV85tbh8w(!cDWLzV1=-^cV?yE|QvqKxz!YLAzsi8!fe=}_s&nnL8lBBK_z6O)&V`**b
zD8v->5+tnF$T4^{(b8N@`m4wC-E9qSsQ@XdpvSoJL!zN07SL!T4=2;`1Z?=kDEOMs
zY6!$Sa?^pXnq!mL8~f~Z#z69%Z)s%gB<@$4sUwo(wK_J<_8%q9bgUs_G{@k=j99bV
zV6q4~?s4sO5X7Y&HJO9lK%@;+VHQhlX(9)j6&qC4v5yN)l#uFk4f?yMq#1$C+X+Ks3E2+27lA9%2MZ%7H>8l}LP|V4~j>
zOYRq2YN`Jg-*TEF$QZ2fuE2CEw2nbjukQBF_o5&|WmwY3CoM84J_Q~=GkaNhQy67J
zW#ssvW{6Nn{*$$~HI%gSQ0DASA}s(d_A+Dm(9KFrfhQNx*J4p~#ZaCkc_=~SGYaN%
zv(S^OjEg0c-jW8tzFF4~u6n(9igdU!(UsEEU3L`m{oIm~P&eibJYX3Kl&}bTxG2H(=7&Khfu#Z@z`=BZnG>Lj41wWF
zh`?z&ppaY^?FDg3+syGq8AgZ889%fzvP%OPfI_ReYE7f1r_7{7BD+n2vVp5N@#$3Z
znu?c2*cvA|JW^Ygv&^P(*X3?7jy@2I#r~}c2xPU;wP}Rf;Ga}jag9=XMbMzxXAg$1
zTQh8D!WK4$SEU$JUXf#PC$^FyFV4a$`ot92I9!)
z&;=l^l%1tE8s&p16>K9L8^e>zb`@yGV3x5=Pnh@-GKO!@T!x&NA}1C`#Si`f?`<%@
zeE&s()A=Nh=v3a{-rUbLNXw)VPXG_9dQs!!Si0_>HiE|&4}c=$57JoAT6sj4+k&=n
zK1O13i=lfLQJ>@u{+|}PNVI*kZx(3^`SRyK{rUCn3)}qiHA(ZJCRF57d
zZH)KxcTZ=UM>fAp^3C_tCzEbUnu`v66ZSYlthV-=HOj-rT`CuhM+EjEzVhA+OqC>I;`>?5WtrdVW2cf@B
zh+-OXxK>KW>==0Ifu>TIGYGGS;x68YL&4P52u(jcO;>b6bam}7$68%am)6+=5IPyZ
zW}IjCUHf(Gz)9v=5UiR(k?AGCq)HAg
z_6;?akkW+{#`Jb#gwX&QIl9Ihen_UtTs_WhVW_mGhYgKlql7qfVvbO!S*S5TqnoTI
zQj5s|82pQK1VA+%M}MZ%>jvQ8i4n&|>0FC_28ulymcX(yub1w9T_B8Y
z&DUJ5K?NK`eCo!8v#ZI7{g*fWiljCU)mgTaB>Fo_cc;+{j7{*lrAQhY?9OuP@yk!W
zM!gR$
z;+xpAwtyMAOhchye_nNjc>IH_Wk$PK0(MVXKtg4V#NFKn1dKyCgs1Yj9pBAfn>au@
zr7ZIZZ{SFT**Rr*%f8-%OqH=ql?;U4V>^EddV)}Bkzt?<5wu%BnG-1%CZHG@5;uMj
zK~6$D9IeraeZ$DtH|uNte0=-mPygwU-#$Kd*b1sI;Q*B`&um=re^I8Nr>XRvKpMq{
z$7m^FB@cv41gyGe@y}p1S>pxuvkRZjo2B$-TUSP+20f
z{KbgDyL8ko+|P3q4R)DoX>$|Fx*=ayKS=?|ny}1n(q)iJZ0jZ5;dg~S_fYjLN>Q(S
zXnKfg3_x044v<+o3Pm%s10535Lv>P8vL$P3%1ENb>XokvelVmWmvvhFTV`3f!IPEI
z6KB#GqV9(_b)`n6(8>f_v%JQi4Uph&p1Qjmz|*V8{j7V3K+(g2l?V~Y5gatWDtoay
zM*NZIxrv$mOa9Nk4WKo;K~OiUPc&uEMO&wwBWa&mZ@kLP19|2htUcxdiT?Yx77fPC
z%ooUy0Vwv6jtLb#MdQZu0hYS8ITJ-`U^!+;XGfX)cm#dc7d0isf7O!7hO}e7LrpZ2
zQH=$4HFrn0E`hEHos_J4oU>`2VsbZeYJ0ZcoTFZ?0&TH@i=OI-RVG3eyhbZ{&vWA+
zJ}K7Rx#NVPkmj5*fipP_w)Df?_a)t|M1msrHF=7%H+0#kLY2j$2WNe
zfeawv!J4=wxxBF*>LuweUT;Q|7KvXkAG!r%j{O5fK6hoYvX7V3%?oADMGH{vJ*!jh
zmcjHMOeiC(mRluyl>PDOYhzzQ=PNFxLfb!Ny6<1Q%T%ehv2dr`aIe0d{Czx+lJZB2
zngTA@4M&CM#*yvQ3}dQHUixzT++2Q^^}HM&pJv>5aJZNXx?DdPj*=Cb_|r%utGipu
z=I(`e@SHw8A;4cf9hMqiwjbW|z~d(dMD-ev);ILO4aZ~T?CU<4C+PaKa(
zeYVrQM|Q3`_>i5Rz`vv-YWU#l&bVHvmaOXA%)ZE|8k$u*WDBn))@R@PxC!
zj*84O71(f_qB?jD&V>P>Zm2k*)(Sks%T$h`3jbUFYZ&s{OeR~c_cMq+
ztMKe0*xiQ+#J4>%RHH7+vHpvgp%w6voH2?>oaq}e&9;f-)D0cI(5HIz|#}ogYZF&(w1_ix!L!xPgTN*Q_q?!nY
z;j2_NMh(B_bPTXFr&2RWQ=a098E1k>YO)pSvl57n+X&N*0d2xmfj-L;59}!Hc(7db
z6L#cI1I4WPuhVY2sllX1ZP;*Gq-)i6x2~|NWFgK4G;*B5U-_nUz1**&1d}T|5$lR&D*>@;IXlxbB@|j^Eh-@~9CW8F1%zLu8#-cP>Yamv)
z!^AU(*$dHC=zQagmIk#%KF}0|<=En=fg$=&`vr`8ky7(@Zy`phv|j@?Xe0RPWDwZA!3@WKe
zE{+MxnilX>=b*Fna2`u>r8fQ2DgF$_Tly6nPK^SWF^z;-&a?;l$QHGdJa0(?Y~=~6
zfgs0DV9}Ww;uFE}g3aKC1tCvurmrIk1tS?@#8+{#RkbXu7?Dk}mgR9@0Y6UZjYX{u
zoSqqIP0i`D{qpA4MB{|49eeWkI+0$20o?V)$Z3(wjlXB*mZCMH%3JGdinXjOO&25k
z+P~kw5s#4rEs0ZeiC}S9d#LIs(h?0~8Gf)E9l)dL(*bdpDrf3!It-^6cuG&}Q(=;<
zfGun~wRUoZ)pUCHLrK|5wcO^Kbzg8^#_Iggt9JGei${h#{M6
zlLJz#VMef-N7#GojMYxT6pX`(x+|Siq^OFOz%D_+d9RMmX*QcL%+r0ABD|NnfBW{$
z${Dw`Y3StZoHm<(`Imq8Kt&VtO7wnGVL-TJVBC5nUu?Aqyuf3K2?(6ZlZ367*_bi5
zgeKF<+q+}xTeotViopaM3@twsWX>NX*neSiFOpu?UU+D7ce?MG2v6NAk9yx7Pdxd2
zJC?|s?gJ)gW;7+w@KE?S`I~BqP%N08)ITbjqjou9px9|YE6I>LV63&xfDAS~I_n!V
z0c|S0{%RJ{6e|OO`F2CN)$r95ON&Tug&B0291&tKXD)eq%g|0<17pD`W^Gu(ozb6F$!=N!48pF?o>6lFDkh-a7_LGm$N%-k|oI!ydJB2
zEO)8QtSSTs5a5A#UJ>DYzzd%mL`DReX`yB4p}H#bE)n5Nto;5*BWL=7u98ZNd)>{=
zRF9th99>MUkjYbo|DcCw&fiZ^gD?EH-k=i8cr3^@nNz^soueY@S9dLI4y;`~->mBYx=
zt}sf2>=0cona$0Gg>+m*n0P_{_1I2|k>RKWV7S{-tX*uBigK9|Z&wM@n8hf1+w4
zR=p(NH|U1Tv=*nzFOZpF@dxg*D{x1)eXqg5R`=%%c+3eKqo0+legDI&U_J&F%%ac?
zs10U=|5OK}7#~U$eefGbWM&1qSXu6qq|$?6uM}mNrLvU07gCEhO&UEc|}K_h03D7a!6u*$=M%nIYGX+={>6pE|acRNfkCNc>>^aq(*
zD-hKq;4Nke`0(5lO@>`uBcH^sQRdmWO3|o@W~RG-%IyAzqSjQz9wo&dk!YzM9ye2u
zTBCz#82ty%ecId@To#ZbV3WRbsxwR=^WriBN{PEP>Pcl|kPqc#fI*l*H@7XDljm1)^dRYU-l|m^LQu(mqv)OX8E7PgYJ!+8sTw
zztYmCo{hWYFW`w6m?1eDz*te2u1oYXf8&H^rb@;+O6>;oDG;`dC}G0@sBSHXk^O6o
z53DN?ADRl+>rZq+jQ@>R#)_#4E;dT(OT7}Hw(=CDryT*%5GNncA`QU-jbLatMkqhU_1u!SIG=bamn?`1vSSXB;zED*eUlOV(h
zZ|5l^ql2Wed1{WashdegO-jwf%G&vDNBrb-NOc@CI$)%MqIy!Gl1-CxRr(Rq@n6p;PPvj%>1W}BN|Yiyga-y>
zZ@JQRR%?N!_j6l0{t$V188YR3olZ04$~A{=Ib34rW#P7b%Y4Pb%^M7-6H(S!*SENLwCzHdOi=EbXbn{
zQFNpw@z{~k=!If33S0B9nkk3+PU_WS$CFTdEd-29e2f=kw(R+f)qfb5gu*>d;AJFd*h
zlbW8%@t$m+kMtac22Hj)MfUZ5dA+~fTwlI^G|M89XCd%QS16m_&S2_y2c*;Vr4XTw
z?3=AD$fv%bQndolT)g$
z1Qe&)S_LUxmC%iP@p1*KrP+L=8tP=0$V!YvS`dqmKnxA#0ijA&-fx6n7uw-e
ztp-7~8Jm4KuNP+Z<@|CjfmC1shGGE-Uh#TeoYU|}4sll9z1|kV0u?_yd1McT6FE=3
z0sP!_5|0$ZCPkg_oJK-z&?KAKD|3FiF92Zr`Pt^f-M98I68}70eq+NZ6M2nt2$&W>
zY|{^WI9V;YF+6}3Ew-zFOEcQ`hZZEY=*Na`IATiCmbpbp>%v0N9nmE9!kpws(mkF!
za>Vl7dFo-Ts?Z2nDxAz?Q;17>;z+!yA+48~wz^&U;aTj)eRem+yg@Fn!eVWrSnz6_XRm+X37(7%(wRjP9%x&GmDhIw}
z3nOLa$Cf9Ycx(#S>##M4LT_YaV<~*1oS5I^3)#TN4!a={Z_?Yo!uGEpO=Bh?ieL1_
z7`aD6f-kLVY8Lg9OZlO7UoIk{fC@NX`RQD797v;v50N*~e%J-?*c1b1m^Ox29Z_I_
zo{3oFt~=Gt!Rmp%yr)K>O;vhiG7g
zGIcdQqJ*7_K*F-NO)CJRDjKj9Qh_OgjnJ4U&_bs18r!j3Zd`J2!CMh>yb&0X356t$
zhXAO%IHYa~YbX!%e9UczZv}fzln$wkwqXX!De;*t#^0Xy#*SW%#M`if9=-lEPw{l~
zbYp0)(Ic8?N?WuS3~5EPRWYQt7b<~u$z_>A3POMK(@Ai|QjnHDB+glSgddrF7Zh3R
zm!KnM?dh7S*(T4=#AAhSqjV#lnyU)(dtg-ezCEgu|lL(7qIV^i)YuydaP_r>@twwMDc|D?M
z*uFt`+8n1ayhO5)eCc@p5DemIwb;FXN&CGg_HUF1I!LsPhIM6WDK$Xtn#O
zoBXPJs^LpV{EutQYmRD6X(#NG4@-h8j-<&W%^5DrzO6>irp`PlAkT*6z#+L1h75EX
zi}IJKDD*t1&dYS45H(M+ph}u9z&O1pPEt;Uq0&K(;FFi5g4vWsQoUx&T;PT6L&?_^
zl=s$A(?;YCa%U!1FX>(xEax$()AjXjnE)`lbfo+4-=s$PCF)m87j8H+hL(nhjh8S_
z;o;k4^fWJ}ErgZ0-|u#}H%1U%b?00?ZXQpcFFYoLmA0Lg4#CD$6-Mt5Eo}$aE&F0%
z6zo#R2yAWEuhceNNdF-K9`|66boqn3glgH3f^?Cp;#m#{IBRIYySKLuD8?1etVH2L)dD`Ef?C*|tp?Q7W@4k`XtMPwh?OwO>;uY78T&h_w
zJKCQKruSFX{VwMrJ?W1nni2vk&_1Wijk*q0770mUFsuG2Hrm-f-iT-?n^bUAlPwk+lw?0SFQ%D7ah;QWX
z9^zO+2z0HYbF(v~Ap)mIGa7UWH8oI1nG$}S1UUrD*26J0U88A>$XqFaCP8^J}m24$04JTd7(1vRA=J4
zb7u)gyAaBaXA?;FCFr+MuW2
zNPjdV;4T-imStAN{aBa>G9=kba)RR02mjQdG-<%YHB)n*eZoVgA+&kodoY~zE{~T7
z(?Z3=VzoL!Yn}dUpZo$qToh)cF}5`BX??G%xITKnc66W=z|vFxEBKB?u=F1W%McTS
z-biv+(Aa$rK087@L4`F1^A?OVgHIcKu0?F!!on9OTEd&dK_mK&LF(su(8{ogy@
zYGgj10po}dv7e`^xFjj8+~}8&X&d}TO=Y9;Bt@rcE_y_=KBB1FKtm2d`~aXr5RN6h
zd=@$Z4RaD5)M~4K()szr!(Og=VZf%5=6cAIu%+iX0D$88292teqv}XfP#(F23BvUN
zBqB~`=|u#@S4ywJet;8YhGiJ|j8W4QOHF?|`Tgyu+wWh${Nq3V!=vG)
z*+BnjTS@-^{BkRVS+P2U>M0%15?+Swo9HM>Ncp5m`G2yOce2eUDKeqD{L%ySEY`f@
zeGXyxD{8eX4)G)r74cc{&amyingB#VyT2l>M40AcRM@|~c{fp0{E0h`jykv4w^SIj
zVI$Pfdu(dRO22d+%UXxv$V?OjTT)ghO1oJxrj5Oi{&ppQ4y#JG3jH|`l=gW)5MmmO
zfoxGIf@l$LdEYq(_?@Gk7eWj
z*aGB!XPl6v8`R(~Bu^H==aHWzO=7f=~rYo}?C-d7d{-$G^s*@GQRpd{Kjl&^!HGs@t)=wg!!U
zvvw=Bln&WJEBKfs+X^vIQ)`i;wHzSGBx+3^Q
zh78kXhs>4!c&MEINYuam>HmFs`1XAGa&!K=-~R$t;~*vtjdVP{c%;=j(K=}ijClE#
zw3X*_fh9H1pum0w
z4sA}dc3kt@&Gb`wABl_qtJeD9I9$vEP&_(61uIENLRPh~Nk>ftrA|FMPm5gyA!~|P
zez8;fK)3LlX1Ihx2EYuo&QKF@SpEvXDySVuKE{&aF*wk0#cDVqG`o$Waolq?OsKz)
z=^!Eq|A5dIl&du8g>C_|v9%~}4PdP4uC@o45q~%PLu=n0x{r7!vd@LABw(smETz`W
z@KcF7t(Agwp=@A-Hav{|RA4G5r{HIbtO)hmX$S_NeWZ|?V-TmEFo~yd5A5drRCd1g
z@P+4J0(oZ5+|`BqFLy^f$9n0&^Y-Gz5`gC5FRtERx5t;A=3CAd3A~0EMNsql1Du_R
z82pYHIP}k_6k{BgE7n?qA5zesXbtfukE4iZ{!fJIm;zHE^rs9*kC04#q>o0C80*qg
z`(P5F)A+AmJV)xBkM0fq@gMx5J#MGY!##R3UWODitUBwoGNl{yruiW}G7f$^)fgE0
zU^`Wl+sy*<0M@+ggGKO%8-7)M;Sc2kkp6DQIeaP9piHvGrTBA|5tX;NOcS-l?Q~^g
zTbNe@C5y2&!t52o!O^QxFgHb?5YKgzOZr5rR;VMtcoKQ7
zZN0RB)s7G|0>P7mhBwpCQppi;twTMyAW@%50Gv<+$CPsidgEVk!H;kV7NT@ydv3JPaKn5Z-6-sT?b9~Fy7@n+Oo{e_>;ll9cIti
zTP$}TK0*}O?8qx)GMMFFVq|xN-1L`ZCZ>;?xfjubyamAKKS{R4mCqSD7ICl+6X35;
za{io!PN%~bkmW}5{y1HS5J^3@D!o*zzFam`>V}cZz#)mIzt1_?_PjOL(Yu6)f)r>Z
zlG3FLr!Uxa0
z;CCvMYnxA)L^&k8nGYQw#<{86EXstYfsjooQG73*N7G!$^)ctic|2X^k#Dat4nCc1
zPB-{&Nxo>F4y%=QOaixg(`7VKojs*4$jB4N#`5t*0Tghd4vWggeQmW+ri*#QrT;nJ
zR9hkKQmscJcj{eTlZCgN0c_eU9mFwJhjL8W+O+M090c#9c0uRz|O>~;?qmtHMQI!%YZJndR-Zh?T|N84H+b}YyH
zUA^=m|JaycjOm#E^%&e}UeQu6A)x33%Z@$ydmwdAiPkkWQa*n)2vtvJ>w#dee7)a)
z`{$RJhka9dw!FWyYYzM|q*Sc^D(Kr*rzC@HP|^lv^*{AV>UyN242u~HCH1>Z#B8o^
zxncz#c6VOFb7E7>d7GQAi2|twal
z;(`LH71ziW4^g3-{^{{>q#;Bpo4z01!j=F+4CbY3
zo)0elgy5{E8H_4S`6>ZgD29NlQ2X_KrA01we>Gaqup+r5#gv6)V^LbGpICYkOl`!U
z_~fjZvf6P|n{Ewut>T?H9q#E&+m4QY26_qsl7Aak7j=TFXvLo;@hwFHtM{L6u6Fl(
z&zGNB7wHM{=c8dQyCQjXPUb0oSdXAIBzOE9kVKvN!JPy~puje^WSq%pkOq*_2rs6G
z1w0T_4m`aM3*zfC$AzvBd7-=*?^=;iY75Zh&7R82gP(2MZYd?sE#uw5$EcDOw$lP)
zvjp*HO~feK9q!W=O*i!E=yAnS6}N}SxqF@9?o}|pwV}cs$Ui>Yv
zlBvIR;@z(ix)_BasctzN2dVr(PWUPpW9VfHj8Yx2b+t(@!b&ZaAN}`jghmg&@=&N5
z>aKMQ*kEKFNWuJrItzbEGw)d{ACXlV=*_{z5x?MXe0JcEp}<&9&~UhWs&mz0n@rNQ
z;QQ|%dH>BLwv(P1Gpby8ytoi@j4c8n=aN)~hd#X~5JsY{?Mz$npIB5$jT$Ih(sr3A
z8{GF@ON;)al_dilhebO2}6u>yM&Hl4yq79l3N@?T0Ixp
zMc{{9KmbWZK~%WkiNM;y
zL{uChYSWBkEA|TdJqCO^?=gV`=wTl^IIBfeK!AqgC`|KY>WmUbW7DcFB&ch{-jgpR
zN-vZ#$W0l*3~X5i26EfDX?QAiG3kFI;lNcX6n*J-jb{!+XA@9$i20uqF^}QQ{#PTP
z21}M5rPUN*2EW94^i5?+Y|qP%1<_QK5KOS(Ue^lInQVS(#y#au#v|)0JJ2;ACuRWE
z3b(|x%oayu1H^m{{9rD0{p=>W#4!swQUg>qQxRAi(AGe$`!&)Sd$Sl+kX~vx9x(x{
zNF7?eK7MnST31Q|9hjD8G_3*X3I^eceY_wNU1QK$@}tLbUw9xP~ZS;spqU`
zf-S+yQ_FFOar|d16;19$wLfj~Ei(0$Dri2I)jupP(IPX1)$7}V9R>dB_3FlWLVH#U
zEK(rLV~jP43q>+6`>pK{EoVWkwSWHp@P|KmFPLtFROI3GfsbX0?Od*1iv!D0@@!lf
z_|kS*jnj`U!wZmxQ4nJSI#kqxkY2d-0}m+XrtYfq#+SBnF)Y3F-?rD+
z-+un{_WJSKo2&*s{MFj+&Bvt!so(NZ2_PjblN&7{XIA|1bl_PYzJ9$wej~%Htzmqt
ziCSYu0r7Z$aemzG?tcE$?*7joZZ`M(?{=hVQ>V8B`KM9kK@y>wDy&SPZVAot>H5n1
ztwcq`_0#i?-fZ1a45^7-n+HPxdxg(f(&82rf${OXvqQLXed
zN>V8&qi<7n%(}X%|NKuuwX7`nLF_ZC2Nv}E*^pK>b0f7{GwGZEVdATR(XU9Zh>LP}
zwz0AhdM8?O>);t^nA`%+nsh%i0dMRicQX(I)-)71StQmXBIyVw^4tO_Un+&mirfu5
zoC;t6)rv5P7vkJdU4bBv0j?5qpGtz_dKlz$6fQ882USaZE>KHJ=>*3ff#^88W8uQ%MrO^782tzmUh
z#4&A3phj1;H1Q^5Q+C8Xms&>>coQ5uBi@!{Fk
zi~u-!1l5m_Bwj#3m25A*1TMQRiYOYQ{Z|?=-{tEe+)?nthk@k2p9lj3frERBHoVP}
zjH#t2Qi6f2nG(2KOGq#dR9N~A@luo(7-kG!%O!N-KjOn{Rc0V;YQPp*hJykTkLoU&cfWfuXPK0tT3}_x-G^Pox&FtPUhVN+MmuTuSZ3R
zEv>3KlAsvHL>I2E?Kt3hj%OV^5l9v92;m*^*jT5z$gHLqi*+JTtB2H@B+nAht(BdT
zfxs-a&2b7{Bd~^7^DK?Fk|-_4@zQ43u@*H>i|l1XdR@c|{tcR6!^sD4&)s
zE8An2K1%UxZTbU1r5IxinwM#SS`WW!*vz=Rm%9#uhgy42Xg
z;1lojr%;d6jx>S27=K5hT;cx%fChZdm$5-MTPL8I0s2YJ
zbFJ)(NKFCd1x&%dzO*RoY_lma3$0
zRHaUAJ0vWXw^9Bg06&pgl9O)`ZQBUdvc!}nNi#Z&m!)IT64-eI0K|vmO*>Z6qKn%q{1*^s$
zoXI(0gfW>m&Cm`s#$DHn0^oHGrarhE7(Vv-(;Qhw)q_=KmGfY8&nao7&yUKF{UIg$;
z+>l2J-J<{a%(^+&nlece*Rs&OssH0?>c0s*`G-M0do5E2rbSsaRzDfmbOnP{0&Ai_
ztfujPrZ?{4qCzD7PL5mjO-;f}rPIU9%k7I~PC?5jY!Y>B>qaiia0j&>WiVoUK&hU0
zP7-hc5XD6q3vae8Ez7zzlU+=`3^2aa#;!|1?A~~CIE3ii>+a(6r=IP{
zO^@tX1U#0$^6)pWSw)4V@YgdgZCH!}+}vItXViwR9-j7x9WSfA^t5$Kq@7#H%^Kt0
z$CFe&oWj??Y+
z2OXsQ`@6HJBlbP+4#)jhTJy_;84o>?M}qLgDq=9YxX{8*j|o1V+7ze>V9S#XHxZ~=b!#do_4ps3s{im+gp!!{K
zt||7~Lx8Cu6>0Rr4Hb|}tD456i+ZC1fkNZMMmbl(3^bKaL)}dGx)KjXzqD&~lTcD|3jA9wZy;{R9AhJ~Be9dVW45yIHm6|^9
zc3!+-mIrQ_@N~T2Uflft!^fY$cn!nj#m$-I7Vgwn$9;Qo{>P7>yx6Tdxf$WwdsJg2
zG3+W}Hw;$iyoB@}{}d=nAWSrgDEX!^F`JfX)gpbj`KIiqvM`
z66R8jScX8`sJ3pC62|}__#L7Lf4DvF(w%vZqTz5($b*(dimQ3Q`6#~(t*{Sq4z0kD
zfYQy8R1fN=|C|+zCr18?zuMjQThRpH;BLtn{!Ak3QgH@KO#oNA3{-IqXT(^^
zzR?!c5Ew^SC&4t7M%5hS2>#zZO0$fSaC@`Y%H#e}e_*uSD(4w*%Es$Z6ssu)RT`v*
z2HC$RD6RA;4hmG{)uatMqBI0sg`S%)Kl5{tdlUR%Fv)u6H<7Q?!bJkuN5u=B{%a+Q77QKdZAm
z23)}meDcEw7&as<0aRa|=4zk^Y(J@yA#S(QAQw|pkYa?^MiauFbVt^fXVx0CNI#iv
z^K+!%Ql52WU+Iz(L5O|nVW8u{^xL9>dD*epdMEQ4l5#!$=Y|4%hU{n!BrXxcljojv
z>d!rUr8;a`Xc2`{1=hhO1#Q?ZYRU}K(^|=`Ag{qVFkuUZo7*T&u;Ad_M5tjNEDDF3
zL9T;iZuL(7)Cb`}e;dlbzVU`jph_HC30b?W6
zD9fMrsA7Jkix5CeY-0-T3*9m0BDR~*DKQ(u%CPX!KumKIDVc}0|L##FxjyI>9^6D9>V4NB2Q|lP`%lrW9d~1t8bc{CfU3}
zp=l*jv)@S<3JX9>-5Y~*!v$?*iv@~~I)~CNBBT<@Lyt9&N9XHT$m()%mOE1=R5%9(
zEdj$%t8*!4jVNsOL9>n%_Io&lERDEd0%e8Fsi`WR*yiHo(>*LA3VHB0Ra>TxIXxH=
z4rK&bHMS9J@Y8%wy4jByn%ktKv-~X9tg$+`y%54s1;c{Cu^sIUg_qw(lmYcz0qiqp
z&V@aG`uK^R1RJ{JolttP$BGX9$<@UNTUzZ6`N_i7yWL!QUWHRP)IyL3Ut&v$YC(dV
zE)6d=fRZvCQpfGr^JlJ7(>BJze?0OEq|NK*Szdg(`*8c=Nh@XWGV~@+p!e;|Uq0)v
z9c-y}vi+kzEfQmhux`!}oAQ0h%|F!o?_u0i59T}{cNU5r9)5m4{PJ-3%W?k=pYJ^D
zee;Pih$|;g()W9LlqXG68#U)mT^^&PRwUlCC_J&Th(*c0UN-~ObiN-yv;3KSZgWo5
zr|l+pO`?n|G|xslj|ToZcdzN#AC8x|5~1d1F+_!kfO^Et88a9wivJmiQlMr~GjPzo
z90eh9%Uw&L?iT!+vx4={DGmjXxyYoB7*q=bXXH|8$~g_VfOEsYIvAAKjEPyEqIOhO
zv^NbEpXO8&kxKR}urOwZ4baXp#4Px3;go@9JNtywiRr%PlOXzP`Zq6@?5Hy|k(1|z
z(;_+Gb1R(Ef&drdTu)BjV`S6De7F(6lp-#L!o1?0cd+Nt_n?vV3LY-!11ediAml2i
zo;HMpXkd#9xEm_{rn?d_s8TVt+%^D3(-mrjiFnZyk5b8ejFxo-TVsZ|0;&rh(>}d8
zN}>&?Z)|#|1C5EFdK~2A_U!cY=j+cmmj|n`j=Rmx4f{R2-Te9IuRnhN^k00~+F020
z=#Q7)^t6{E&;)9QE&T09fIPFhz8|x(tI$scIt;1gtYeJHv?v*&C4;;iX%o`$Cz?v+
z6fL7GexgnE=jbwP!(0;ck|-SVKowZcAQ&3#l~n0$F*ymwRVQ(B*jQ$R`K=R8>*6i%
zJ%c3L>Vg(bB;lxM|4JaHaZQv$0@sitF`s&>CVhdtfjwhkVHOpe|(vkf(X-v-4<7+Z*X7=7kA#aY>6jL1m1i8+Y|u@RDYvL|mgSr*3qOOJYFDP*XO!i2T`EV@F~!>j-~nD~$&=&hrAOPj)~EJzZXH+g
zr)1sea`erp+cG-55VKROOWT=O44z55^|$qt`KTeE?ZrQAnV*KY8x!Rfr;C9WT3tUQ
zk4cz(c{y_l-poTv(O--7@mGlyy1Y$!gIJ9Uc;PwH8_st(k;ewwKQ`-{hCMXGg7M5Y
zb{Z`WHf@D6tkyE@VuW2>M<)q=`HhzB$B$G`(C0d}iqCYFMp`Fco4@VAbx`Y34}46e
zlPP-q?HaHXq_v2=G{I#@yyFLlAvseQ2fqqrHm;1Q2+tj%;n;#ewi3GmS1N_beg=MR
zk_CW?Mp53X&P9!l(|LzqcHyn=-Hi&Pf&)t`A
zAAbh|`M;p@x(=i~QphwooaKuOvq_K&m2z;5^T
zQj)*~yl?j4vGo#e5m^A$cf)4;G?m!AJc@ZtXn%}R$+A8B;{i_dYgD|{{GxXnx~ekz
zNi5wRZNMq!&z^CJT~tq8}xFx1G^@q-05Gaxco}JDX1{Q|U!9uBMK-
z00DTXen7$kM5Q|M9~5O6Y34Yj
zhnK5S8ip>WKZ8)0C-hEBpu0auhPI&jS$qBhGEJjKZEggpVsu0f63R>gKmzUJ8+V*4gA3>D1+hz2zSjAm%+W-=was_<}%RJv)dc*>wCKK48?ZSqORT3=g&Vr
zK7IQ6>v!4r)z!&=|KI-c}$JCPzVU
z@{LY8roq^!JNXfAg@oYQl|@mAzLx^KtT)R;i=KLXQE?ss$L-{bk
zQ=G$jS^?IQa2KjCJ}KC8l9Ur$LbYPUs_tiR{fcQzs|HxtL&?3An@BoB7z$AinJlX`
z9_TEfTW$<^KNjS02$5+MQmgqnNo}VRo|a+x3C`4vjK~IS;CJz_93uiFH=)7kSitkq
zE(fxcD)HB$x7zj_DsaZvboTZ7a(m3L7k8FGAx_dW?JwD}vqk|*6);NOD3Xj`*pSLA
zDFn%NV1S3frHF8@d&kibtK`;Koj}s8;S`X15pSF_9wK|2Jj46RQ9riTG7^}^wwP30
z)XS*6i&|iN+LOx4iIk#t^D_Z`A8ni@R#!&?9zT7W
z)Ves?UftdZYIdDD^FqXSJ!|_I!(gwMwzjk5*ep!Bi!@fw2MUl
zbsq=fkIVvAnG_9cOIbZTBn84Z97_kD3(e?<&y>x-F=IxmRE9p#Nw`HGJ=e}h8~@W5
zCQzl3wtNFlfsP-sr}RDel8JDg;WC^`r4n$wD_4eJo~UcS6C_juTL-WXU7}P*X_RFp
zbx@CHFwmqB6w?7Mgl)yDpV)>#G65qUR$xe1H2bBYAx>p+Dovp|BNW=iL$RoMQY9AZ
zW*!6>oP~!wKH%@rYW)adG^fqciS*D#6l2FB8SM*fnsW{OStPQ%f0IIaD20}96
zKs$q?FgSj)TYvw7kKqtkf|ieQ1`jlcS#H~pSg(NpF~v!KjUJRP#J`Y9B4`{1Pki`3
zx6%{c^e_$#03-Pv%&=zsOpJ|tu^x!TIp`$nek2Ou%7sH@Ag8mdPFQyqTS#SeBb4Yr
z{;QC?&^*N$CdNGQC82HyTA5rJ7HcpU)2~YT=@R>hE`41(;DJ@Gj8HvBiAwsA%{idg
zv`eaOyGIG02ptz23X!Z+0En=j;h`4d85fG0H56EN@f4(?;>1QMvt@w=meB$15sMjwU3$S+;FwqS!O>g3hJBM17L89u_JyF@QsE$3tdSupnf@
zk=!Ss>B~dM5Dz4*Zw&PmIm~)~{dm~*Ecxd1$)zn`pZ@sM@2@|79Q9BR15!m15E{{srnDbmsC*51D
z>Prk+KimX(>9TVO{0sQblUS2;6dK~F8;!+{z@k3*Rq7;UQ|Hjn^mAP=()Sa~R~he(
z0a7#Kr{{Nv!-vl|w3Wc&-BHhRqj|F
z>h_KdLU|c~xpV-B9hyi}LN4@&W$$|AP9w_TLyGEzr4UjDg=fI2Dq$C>FdPL_Jx%XM
z|D4t8`n~k1$A=@JI>Sdp+^a~%{@RhDiweR>K@JEIIC^PBQi0>L1>Bt$qo>iGXM?;N
z(9d)j`Ow5LDVGcZnHs^AUbsP$68FYRV~H%v0on{ik@cV*vN(vNNLB~?GwTK{CvFW1
z<*(%j__5&PVtG%fmUh3rTDjrw0=2lvxZ~yG4g|Os`Y&-E)z?taC=Z#}xITv5YKZwY
z424&EHbujeO{c7!y^|7|AXFU`KkRy*M=q@@maR5XMe7|;Cv7tvvoNT{46^hyg>zc*
z?sZa23`9OYJB?zR>M3(8%V$)@$)~Wo>@ui4A-1t
zjkuvMt|bmlG)On+ezGG(f&qWJrL(1CRQpoF`?ZVcE{FuFmuAg@Y-uQkotf@=sIjNa
zhpW(R&`eGeMslz_W43
zZjIy5Gp^f-eoCc3akx+FG$~C$fFeB-05U~s?GelDdXh~O#si^*yhd{M=zt?%dXxVL
zu-mS=lZ9zK6DB=en4DXDF})cb5QrFpHEaqObzPLgRT+E~sKrMaO9me@St*g%St0x~
z$RwUkF>lvGJC*4bt=kM`YnsuL3)k9HoXUZ#%Z?_g{J5Rf8^O>u*PrbnY
zDfxBiZUxw@Y;nV=6_SbP{G}-}v#`B1O~QY67)MAENLjgAqe~y(45)G}Sh=!6ED=X2
zA(+yY$HV>I)$QkRU%zZ`O)0cEC-Icy%@Na^!O3M8XtE_Y)dxDDEIDz>pu!Xs|+QNYnGIyeYj;(ucvLs?v7=Ik7%B2i?H5)#-
zH|FfjmuAGu_rakc;NO)O`gzcN6NQ2^N^^6l)yX1m{i{o$t{&yKyh2{cbtfBJaS
z(gaDBLvpwzmFsf4d3kPhgNaI7LzQWRW-0r5ztf`Vyx~4Lo
zoGuTS3rnosoYwmpM94(9s}1ww5k5?Ob~BqOV6TOV7zeNR<(E%{10>|)C@ZRFgHQl0
zWMW}v>fF!;=*rKB7D~v{38)3q71-rLmgaS3{&gpYs17NY;m?e4Y($n9)?{V%$Ku8!
zK4pwK_$jXn`rkS-E~Ftfb*dqQd-89fo$GHHY=MB*Kn09rm3mldgS)L{C=bEBR+L6&
zsFD$iq!$fC`jtN_bInNO6BTx_>#;#Gv!>F5m
zMO?aEG-1=Oj)xuMtEfgtH6bxFNEK2vI=SwK&$xj&mqw>G!%wC7l#~hr3a4UKH~3km
z8@SOC@^;k{katI7{@kMLD&i)Il|Kw@z)x(k8C#JCkJQsGM8venOC_=#VJh`-A)RdO
z+}!ahf;!{7)iNS$G|s@%#l^w447QUKeR+{aqe@=9)dflru=TzpU8g_z9CA@tQY)Vx_oG
z4l9=uGkv-XO#Jo3KhW7bjnpzdTtwykz>Z?QmJzRV{<+~>QC}KNgQj)Ax}ZpnX9uCP
z(9g0pU052Xg>xQ%1B+@o?
zlHk;FQ;BSSuC$R>B?&*!|M^J1RWNhiK(QJTlq}=17Siwq5vcd7v4ou9Ie=0opndU9
z!>`nJwLvFT2iFJ%CzLG>(rpf^I7x3#@9y_k*Ecs;b}AiJ=Gjv(2(N>aUcYsrHUPtw~ki%e>pZR)}mlpze+mzc)LDQ(W+s3I9b^C&5C!
zd);(mK(du$;NnK{u%2tp7q?8IIwE)4hhUue+A1PZ9cX&e^Uthx^>
zOX{~{F&@y4(T-Jqq?TfgHsp2YMDZG-H#s5Eg*taF$DGrZwyCf1P_F2&;t%h}C0s{p
zUfZ}KA-%iMkpzaR94ZzlO;Hir889obr#&H{#n=Ux`9*$2EQUSLKLD=6!bCtwo~4&Acuh~N|(
zDsz-Y&ZyT6Un&ItjNIT1Vh6$MIM_*qCciYRRVBnq8ASRl3(
zQk-dYIWdATeI~3MLVQA1b)LFa%ehA7QgK2}Lu)sVOE8E9hIK=lzG*0~h+BX(WCEQC
zuLY<~5}{1LRQ1w6-LD48g#oh|;}F~$goptf%gn-&zzSKi+r&eF#c%W4AcGhDTvdy6
zQYK}(?!ys*EfQP$*&xAExX}}ghB+N*M`xo%Gcm$Gg-p}xuP?`&+uQr^Hfk`i*CwHD
z?9*Dg;RdoNa{EZVn3TqK9?D4jkxn+bmQ-BA5SZj%@yey~^ocy1AlmfX#eCmLHSzT$
zClF6G;oA_H=45E}IH_IQp-2OH&IAK8-jYDNr$+i8E~UvX0+H$@J%9b1|M6x96J9oG
zC`IUoOJ?e8R72!2QX^lM?e2&*PK+)B3)h+sCUhwEj5vgU>mO8TrM~ZB{KCQI`~xzn
zLuqG6PiRd0nP7~r$}n^B$IrU45iA;5xXXP*?(%$_GZtT3E|GfFTufQ2yG;^arrS5M#IBvba_%h2&<6%qlsZw~OdL5yTzPPv4plntV
z*F!M<>9+zSDyK5#ay