.gitattributes

@@ -1,2 +0,0 @@
-# Auto detect text files and perform LF normalization
-* text=auto

.gitignore

@@ -1,6 +0,0 @@
-/.vscode/
-*.pt
-/.idea/
-
-__pycache__/*
-*/__pycache__/*

.idea/.gitignore

@@ -0,0 +1,8 @@
+# Default ignored files
+/shelf/
+/workspace.xml
+# Editor-based HTTP Client requests
+/httpRequests/
+# Datasource local storage ignored files
+/dataSources/
+/dataSources.local.xml

.idea/deployment.xml

@@ -0,0 +1,22 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<project version="4">
+  <component name="PublishConfigData" autoUpload="Always" exclude=".svn;.cvs;.idea;.DS_Store;.git;.hg;*.hprof;*.pyc;test_imgs" remoteFilesAllowedToDisappearOnAutoupload="false">
+    <serverData>
+      <paths name="root@192.168.199.249:4868">
+        <serverdata>
+          <mappings>
+            <mapping deploy="/workspace" local="$PROJECT_DIR$" />
+          </mappings>
+        </serverdata>
+      </paths>
+      <paths name="root@192.168.199.249:4868 (2)">
+        <serverdata>
+          <mappings>
+            <mapping deploy="/workspace" local="$PROJECT_DIR$" />
+          </mappings>
+        </serverdata>
+      </paths>
+    </serverData>
+    <option name="myAutoUpload" value="ALWAYS" />
+  </component>
+</project>

.idea/inspectionProfiles/profiles_settings.xml

@@ -0,0 +1,6 @@
+<component name="InspectionProjectProfileManager">
+  <settings>
+    <option name="USE_PROJECT_PROFILE" value="false" />
+    <version value="1.0" />
+  </settings>
+</component>

.idea/misc.xml

@@ -0,0 +1,4 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<project version="4">
+  <component name="ProjectRootManager" version="2" project-jdk-name="Python 3.8 (py38)" project-jdk-type="Python SDK" />
+</project>

.idea/modules.xml

@@ -0,0 +1,8 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<project version="4">
+  <component name="ProjectModuleManager">
+    <modules>
+      <module fileurl="file://$PROJECT_DIR$/.idea/yl-ocr-layout.iml" filepath="$PROJECT_DIR$/.idea/yl-ocr-layout.iml" />
+    </modules>
+  </component>
+</project>

.idea/vcs.xml

@@ -0,0 +1,6 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<project version="4">
+  <component name="VcsDirectoryMappings">
+    <mapping directory="$PROJECT_DIR$" vcs="Git" />
+  </component>
+</project>

.idea/yl-ocr-layout.iml

@@ -0,0 +1,20 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<module type="PYTHON_MODULE" version="4">
+  <component name="NewModuleRootManager">
+    <content url="file://$MODULE_DIR$" />
+    <orderEntry type="jdk" jdkName="Python 3.8 (py38)" jdkType="Python SDK" />
+    <orderEntry type="sourceFolder" forTests="false" />
+  </component>
+  <component name="PyDocumentationSettings">
+    <option name="format" value="PLAIN" />
+    <option name="myDocStringFormat" value="Plain" />
+  </component>
+  <component name="TemplatesService">
+    <option name="TEMPLATE_CONFIGURATION" value="Jinja2" />
+    <option name="TEMPLATE_FOLDERS">
+      <list>
+        <option value="$MODULE_DIR$/templates" />
+      </list>
+    </option>
+  </component>
+</module>

Dockerfile

@@ -93,7 +93,7 @@ startretries=0\n\
 redirect_stderr=true\n\
 stdout_logfile=/var/log/be.log\n\
 stdout_logfile_maxbytes=50MB\n\
-environment=CUDA_VISIBLE_DEVICES=0,PYTHONUNBUFFERED=1,PYTHONPATH=/worksapce/yolov5\n\
+environment=CUDA_VISIBLE_DEVICES=0,PYTHONPATH=/worksapce/yolov5\n\
 " > /etc/supervisor/conf.d/be.conf
 
 # envs/py38/

client.py

@@ -13,8 +13,8 @@ from io import BytesIO
 from PIL import Image
 
 def send_request(file_list = ['./images/zidane.jpg'], 
-                    model_name = 'yolov5x',
-                    img_size = 800,
+                    model_name = 'yolov5s',
+                    img_size = 640,
                     download_image = False):
 
     #upload multiple files as list of tuples

create_md.py

@@ -0,0 +1,182 @@
+"""
+function: 生成md文件
+"""
+from mdutils.mdutils import MdUtils
+import os
+import cv2
+import numpy as np
+import requests
+
+d_map = {0: 'code', 1: 'logo', 2: 'style', 3: 'table', 4: 'text', 5: 'title'}
+images_path = "./train/images/"
+labels_path = "./train/labels/"
+cut_ori_img_path = "./train/cut_ori_imgs/"
+cut_predict_img_path = "./train/cut_predict_imgs/"
+md_cut_ori_img_path = "../train/cut_ori_imgs/"
+md_cut_predict_img_path = "../train/cut_predict_imgs/"
+md_path = "mdfiles/"
+
+
+# 标签是经过归一化的，需要变回来
+# xywh格式 ---> box四个顶点的坐标
+def xywh2lrbt(img_w, img_h, box):
+    c, x, y, w, h = int(box[0]), float(box[1]), float(box[2]), float(box[3]), float(box[4])
+    x = x * img_w  # 中心坐标x
+    w = w * img_w  # box的宽
+    y = y * img_h  # 中心坐标y
+    h = h * img_h  # box的高
+
+    lt_x, lt_y = x - w / 2, y - h / 2  # left_top_x, left_top_y
+    lb_x, lb_y = x - w / 2, y + h / 2  # left_bottom_x, left_bottom_y
+    rb_x, rb_y = x + w / 2, y + h / 2  # right_bottom_x, right_bottom_y
+    rt_x, rt_y = x + w / 2, y - h / 2  # right_top_x, right_bottom_y
+
+    lrbt = [[lt_x, lt_y], [lb_x, lb_y], [rb_x, rb_y], [rt_x, rt_y]]
+
+    return lrbt, c
+
+
+def IOU(rect1, rect2):
+    xmin1, ymin1, xmax1, ymax1 = rect1
+    xmin2, ymin2, xmax2, ymax2 = rect2
+    s1 = (xmax1 - xmin1) * (ymax1 - ymin1)
+    s2 = (xmax2 - xmin2) * (ymax2 - ymin2)
+
+    sum_area = s1 + s2
+
+    left = max(xmin2, xmin1)
+    right = min(xmax2, xmax1)
+    top = max(ymin2, ymin1)
+    bottom = min(ymax2, ymax1)
+
+    if left >= right or top >= bottom:
+        return 0
+
+    intersection = (right - left) * (bottom - top)
+    return intersection / (sum_area - intersection) * 1.0
+
+
+def send_requests(img_path):
+    b_img_list = []
+    b_img_list.append(('file_list', ('image.jpeg', open(img_path, 'rb'), 'image/jpeg')))
+    payload = {'model_name': 'ocr-layout', 'img_size': 1920, 'download_image': False}
+    response = requests.request("POST", "http://192.168.199.249:4869/detect", headers={}, data=payload,
+                                files=b_img_list)
+    result = eval(str(response.text))
+    result = eval(str(result['result']))
+    return result
+
+
+if __name__ == "__main__":
+    list_images = os.listdir(images_path)
+    list_labels = os.listdir(labels_path)
+    all_acc = 0
+    all_label_len = 0
+
+    # 遍历每张图片
+    for m in range(len(list_images)):
+        each_acc_map = {0: 0, 1: 0, 2: 0, 3: 0, 4: 0, 5: 0}  # 每张图片对应的各自预测正确数量
+        each_acc = 0  # 每张图片总共预测正确的数量
+
+        image = list_images[m]
+        img_name = image.split(".")[0]
+        # if image != "125-------------_jpg.rf.9a0d958c405ffb596b4022e13f0686b7.jpg":
+        #     continue
+        for n in range(len(list_labels)):
+            label = list_labels[n]
+            label_name = label.split(".")[0]
+            if img_name != label_name:
+                continue
+            if img_name == label_name:
+                print(image)
+                print(label)
+                print("===========")
+                # 每张图都创建一个md文件
+                mdFile = MdUtils(file_name=md_path+'yq_0819_' + str(img_name), title='yq_0819_' + str(img_name) + '_iou>=0.5')
+                list_item = ["ori_cla", "predict_cla", "cut_ori_img", "cut_predict_img", "isTrue"]
+
+                result = send_requests(images_path + image)
+
+                img = cv2.imread(images_path + image)
+                img_h, img_w, _ = img.shape
+                # 获取原图上的切割结果
+                arr_label = []  # labels的边界框，存储四个顶点坐标
+                ori_clas = []  # label的真实类别
+                with open(labels_path + label, mode="r", encoding="utf-8") as f:
+                    lines = f.readlines()
+                for line in lines:
+                    line = line.split(" ")
+                    lrbt, c = xywh2lrbt(img_w, img_h, line)
+                    arr_label.append(lrbt)
+                    ori_clas.append(c)
+
+                all_label_len += len(lines)  # 所有标签的数量，为了后面计算所有图片的准确度
+
+                num = 0
+                for i in range(len(arr_label)):
+                    lt_x, lt_y = int(arr_label[i][0][0]), int(arr_label[i][0][1])
+                    rb_x, rb_y = int(arr_label[i][2][0]), int(arr_label[i][2][1])
+                    rect1 = [lt_x, lt_y, rb_x, rb_y]
+
+                    # 切割真实标签，并保存到本地
+                    cut_ori_img = img[lt_y:rb_y, lt_x:rb_x, :]
+                    save_cut_ori_img = cut_ori_img_path + str(img_name) + "_" + str(num) + ".jpg"
+                    cv2.imwrite(save_cut_ori_img, cut_ori_img)
+                    save_cut_ori_img = mdFile.new_inline_image(text='', path=md_cut_ori_img_path+ str(img_name) + "_" + str(num) + ".jpg")
+                    ori_cla = int(ori_clas[i])
+                    print("此时的ori_cla:", d_map[ori_cla])
+                    num += 1
+
+                    for j in range(len(result[0])):
+                        res = []
+                        predict_cla = int(result[0][j]['class'])
+                        print("此时的predict_cla:", d_map[predict_cla])
+                        if ori_cla == predict_cla or (ori_cla in [2, 3, 4] and predict_cla in [2, 3, 4]):
+                            rect2 = result[0][j]['bbox']
+                            iou = IOU(rect1, rect2)
+                            if iou >= 0.5:
+                                # 切割预测结果
+                                cut_predict_img = img[rect2[1]:rect2[3], rect2[0]:rect2[2], :]
+                                save_cut_predict_img = cut_predict_img_path + str(img_name) + "_" + str(
+                                    j) + ".jpg"
+                                cv2.imwrite(save_cut_predict_img, cut_predict_img)
+                                save_cut_predict_img = mdFile.new_inline_image(text='', path=md_cut_predict_img_path+ str(img_name) + "_" + str(j) + ".jpg")
+                                isTrue = True
+                                list_item.extend(
+                                    [d_map[ori_cla], d_map[predict_cla], save_cut_ori_img, save_cut_predict_img,
+                                     isTrue])
+                                each_acc_map[ori_cla] += 1
+                                all_acc += 1
+                                each_acc += 1
+                                break
+                        if j == len(result[0]) - 1:
+                            isTrue = False
+                            list_item.extend([d_map[ori_cla], " ", save_cut_ori_img, " ", isTrue])
+                            list_item.extend(res)
+
+                # 每一张图片的总准确度
+                each_acc = round(each_acc / len(ori_clas), 3)
+                # 每一类的准确度
+                code = round(each_acc_map[0] / ori_clas.count(0), 3) if ori_clas.count(0) else "这张图没有这个标签"
+                logo = round(each_acc_map[1] / ori_clas.count(1), 3) if ori_clas.count(1) else "这张图没有这个标签"
+                style = round(each_acc_map[2] / ori_clas.count(2), 3) if ori_clas.count(2) else "这张图没有这个标签"
+                table = round(each_acc_map[3] / ori_clas.count(3), 3) if ori_clas.count(3) else "这张图没有这个标签"
+                text = round(each_acc_map[4] / ori_clas.count(4), 3) if ori_clas.count(4) else "这张图没有这个标签"
+                title = round(each_acc_map[5] / ori_clas.count(5), 3) if ori_clas.count(5) else "这张图没有这个标签"
+
+                mdFile.new_line(str(image) + "—总准确率:" + str(each_acc))
+                mdFile.new_line(str(d_map[0]) + "—准确率:" + str(code))
+                mdFile.new_line(str(d_map[1]) + "—准确率:" + str(logo))
+                mdFile.new_line(str(d_map[2]) + "—准确率:" + str(style))
+                mdFile.new_line(str(d_map[3]) + "—准确率:" + str(table))
+                mdFile.new_line(str(d_map[4]) + "—准确率:" + str(text))
+                mdFile.new_line(str(d_map[5]) + "—准确率:" + str(title))
+
+                break
+
+        mdFile.new_line()
+        mdFile.new_table(columns=5, rows=len(list_item) // 5, text=list_item, text_align='center')
+        mdFile.create_md_file()
+
+    print(all_acc, all_label_len)
+    print("所有图片的预测准确度：", all_acc / all_label_len)

cutimgs.py

@@ -0,0 +1,110 @@
+import cv2
+import os
+import matplotlib.pyplot as plt
+
+path = "./imgs/"
+cut_img_path = "./cut_img/"
+d_map = {0: 'code', 1: 'logo', 2: 'style', 3: 'table', 4: 'text', 5: 'title'}
+
+def makedir(dir):
+    if not os.path.exists(dir):
+        os.makedirs(dir)
+
+def xywh2lrbt(img_w, img_h, box):
+    c, x, y, w, h = int(box[0]), float(box[1]), float(box[2]), float(box[3]), float(box[4])
+    print(c, x, y, w, h)
+    print("+++++++++++++++++")
+    x = x * img_w  # 中心坐标x
+    w = w * img_w  # box的宽
+    y = y * img_h  # 中心坐标y
+    h = h * img_h  # box的高
+    print(c, x, y, w, h)
+
+    lt_x, lt_y = max(0, x - w / 2), max(0, y - h / 2)  # left_top_x, left_top_y
+    lb_x, lb_y = max(0, x - w / 2), min(img_h, y + h / 2)  # left_bottom_x, left_bottom_y
+    rb_x, rb_y = min(img_w, x + w / 2), min(img_h, y + h / 2)  # right_bottom_x, right_bottom_y
+    rt_x, rt_y = min(img_w, x + w / 2), max(0, y - h / 2)  # right_top_x, right_bottom_y
+
+    lrbt = [[lt_x, lt_y], [lb_x, lb_y], [rb_x, rb_y], [rt_x, rt_y]]
+    print(lrbt)
+
+    return lrbt, c
+
+
+img_cls = os.listdir(path)  # 获取train，valid，test目录名称
+print(img_cls)
+# 遍历每个目录，拿到图片和标签
+for img_cla in img_cls:
+    # print(img_cla)
+    images_path = path + img_cla + "/images/"
+    labels_path = path + img_cla + "/labels/"
+    list_images = os.listdir(images_path)
+    list_labels = os.listdir(labels_path)
+
+    for m in range(len(list_images)):
+        image = list_images[m]
+        img_name = os.path.splitext(image)[0]
+        if image != "784_jpg.rf.8a0d3a961866d9bda06a53d9bf2fbc40.jpg":
+            continue
+        for n in range(len(list_labels)):
+            label = list_labels[n]
+            label_name = os.path.splitext(label)[0]
+
+            if img_name != label_name:
+                continue
+            else:
+                # print(image)
+                # print(label)
+                # print(img_cla)
+                # print("==================")
+
+                img = cv2.imread(images_path + image)
+                img_h, img_w, _ = img.shape
+                arr_label = []  # labels的边界框，存储四个顶点坐标
+                ori_clas = []  # label的真实类别
+                with open(labels_path + label, mode="r", encoding="utf-8") as f:
+                    lines = f.readlines()
+
+                # some labels are [],-----> image not idea
+                if not lines:
+                    break
+
+                for line in lines:
+                    line = line.split(" ")
+                    lrbt, c = xywh2lrbt(img_w, img_h, line)
+                    arr_label.append(lrbt)
+                    ori_clas.append(c)
+
+                num = 0
+                for i in range(len(arr_label)):
+                    lt_x, lt_y = int(arr_label[i][0][0]), int(arr_label[i][0][1])
+                    rb_x, rb_y = int(arr_label[i][2][0]), int(arr_label[i][2][1])
+
+                    cls_num = ori_clas[i]
+                    cls = d_map[cls_num]
+                    print(cls)
+                    makedir(cut_img_path + cls)
+
+                    # 切割真实标签，并保存到本地
+                    # print(lt_y)
+                    # print(rb_y)
+                    # print(lt_x)
+                    # print(rb_x)
+                    if lt_x >= rb_x or lt_y >= rb_y:
+                        continue
+                    cut_ori_img = img[lt_y:rb_y, lt_x:rb_x, :]
+                    # plt.imshow(cut_ori_img[:,:,::-1])
+                    # plt.show()
+                    save_cut_ori_img = cut_img_path + cls + "/" + str(img_cla) + "_" + str(m) + "_" + str(cls) + "_" + str(num) + ".jpg"
+                    # cv2.imwrite(save_cut_ori_img, cut_ori_img)
+                    num += 1
+
+                break
+
+
+
+
+
+
+
+

docker-compose.yml

@@ -10,7 +10,8 @@ services:
     tty: true
     working_dir: /workspace
     ports:
-      - '18089:8080'
+      - '4869:8080'
+      - '4868:22'
     # volumes:
     #   - ./:/workspace
     # deploy:

requirements.txt

@@ -36,4 +36,4 @@ pandas
 # extras --------------------------------------
 thop  # FLOPS computation
 pycocotools>=2.0  # COCO mAP
-torchvision>=0.8.1
+torchvision>=0.8.1

server.py

@@ -1,4 +1,3 @@
-# -*- coding:utf-8 -*-
 from fastapi import FastAPI, Request, Form, File, UploadFile
 from fastapi.templating import Jinja2Templates
 from pydantic import BaseModel
@@ -18,6 +17,7 @@ YOLO_DIR = '/workspace/yolov5'
 # WEIGHTS = '/workspace/best.pt'
 # WEIGHTS = '/workspace/yili.pt'
 # WEIGHTS = '/workspace/best2.pt'
+
 WEIGHTS = '/workspace/0822yili.pt'
 
 logger = logging.getLogger('log')
@@ -43,7 +43,7 @@ logger.addHandler(ch)
 
 device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 bl = torch.cuda.is_available()
-#logger.info(f'是否可使用GPU=======>{bl}')
+logger.info(f'是否可使用GPU=======>{bl}')
 
 app = FastAPI()
 templates = Jinja2Templates(directory = 'templates')
@@ -56,7 +56,6 @@ colors = [tuple([random.randint(0, 255) for _ in range(3)]) for _ in range(100)]
 
 if model_dict['ocr-layout'] is None:
           model_dict['ocr-layout'] = model = torch.hub.load(YOLO_DIR, 'custom', path=WEIGHTS, source='local').to(device)
-#          print(model_dict['ocr-layout'])
           logger.info("========>模型加载成功")
 
 
@@ -93,7 +92,7 @@ def drag_and_drop_detect(request: Request):
 def detect_via_web_form(request: Request,
                         file_list: List[UploadFile] = File(...),
                         model_name: str = Form(...),
-                        img_size: int = Form(800)):
+                        img_size: int = Form(1824)):
 
     '''
     Requires an image file upload, model name (ex. yolov5s). Optional image size parameter (Default 1824).
@@ -111,8 +110,7 @@ def detect_via_web_form(request: Request,
     #create a copy that corrects for cv2.imdecode generating BGR images instead of RGB
     #using cvtColor instead of [...,::-1] to keep array contiguous in RAM
     img_batch_rgb = [cv2.cvtColor(img, cv2.COLOR_BGR2RGB) for img in img_batch]
-    print(img_size)
-    print('111111')
+
     results = model_dict[model_name](img_batch_rgb, size = img_size)
 
     json_results = results_to_json(results,model_dict[model_name])
@@ -142,7 +140,7 @@ def detect_via_web_form(request: Request,
 def detect_via_api(request: Request,
                 file_list: List[UploadFile] = File(...),
                 model_name: str = Form(...),
-                img_size: Optional[int] = Form(800),
+                img_size: Optional[int] = Form(1920),
                 download_image: Optional[bool] = Form(False)):
 
     '''
@@ -162,10 +160,7 @@ def detect_via_api(request: Request,
     #create a copy that corrects for cv2.imdecode generating BGR images instead of RGB,
     #using cvtColor instead of [...,::-1] to keep array contiguous in RAM
     img_batch_rgb = [cv2.cvtColor(img, cv2.COLOR_BGR2RGB) for img in img_batch]
-    print('111111')
-    print(img_size)
-    img_size= 800
-    print(img_size)
+
     results = model_dict[model_name](img_batch_rgb, size = img_size)
     json_results = results_to_json(results,model_dict[model_name])
 

test_imgs/labels/01.txt

@@ -0,0 +1,27 @@
+2 0.9015390252839869 0.40270578968097676 0.15897764748992305 0.12198897203623474
+2 0.12520703554415535 0.4054430878298543 0.11080615610113594 0.11905474596297753
+3 0.2789153536093807 0.43694761717211505 0.15605716379626236 0.1442181961402127
+4 0.7265335287651155 0.4378810555336747 0.16112495419567607 0.24911776289877904
+5 0.4996775375595456 0.45831429696731 0.2141223891535361 0.14699094131547852
+2 0.9089739831440089 0.5284245766049626 0.13862953462806887 0.11057502953918866
+2 0.1255661414437523 0.5302796376526192 0.1115243679003298 0.11231193383221742
+1 0.4928032246244045 0.5507640803465932 0.03995236350311469 0.03959826703426546
+1 0.42574935873946496 0.5712721543914927 0.06039941370465372 0.06589602205592753
+4 0.2807621839501649 0.5893619535250099 0.15584096738732137 0.16515951161874753
+1 0.14955294979846098 0.6205947223316266 0.05776474899230487 0.06033871602993304
+1 0.5632979113228289 0.6240055139818826 0.09681201905459876 0.11879086254430879
+4 0.9551264199340418 0.624671130366286 0.03138145840967387 0.03600630169358015
+4 0.09986075485525833 0.6260259944860181 0.0366397947966288 0.03662071681764474
+0 0.7260754855258337 0.6266482867270579 0.1582521069989007 0.12618353682552186
+4 0.9577720776841334 0.6712997243009059 0.02884939538292415 0.04095313115399764
+4 0.12024184683034078 0.6721425758172509 0.07935507511909125 0.04052776683733754
+4 0.42443752290216197 0.6773060259944861 0.09104067423964822 0.03614415124064592
+4 0.8655734701355808 0.6776486805829066 0.08313301575668744 0.024363922804253643
+1 0.21567973616709418 0.1499369830641985 0.05060828142176622 0.06125640015754234
+4 0.782103334554782 0.154883812524616 0.060971051667277384 0.06163450177235132
+4 0.5012165628435324 0.1885191020086648 0.14333455478197143 0.03263489562820008
+4 0.9373506779039942 0.19026782197715636 0.05257237083180652 0.020949192595510042
+1 0.5020520337119824 0.2776092949980307 0.19535360938072555 0.11387554155179204
+1 0.1301209234151704 0.27997243009058687 0.10836203737632834 0.10729814887751084
+2 0.903081714913888 0.28357227254824735 0.15814950531330157 0.10031902323749509
+4 0.25807255404910223 0.3398227648680583 0.11633198973983146 0.03618353682552186

test_imgs/labels/02.txt

@@ -0,0 +1,27 @@
+2 0.9015390252839869 0.40270578968097676 0.15897764748992305 0.12198897203623474
+2 0.1271528032246244 0.40813706183536824 0.11469769146207402 0.12444269397400551
+4 0.7261744228655185 0.44918077983458055 0.16040674239648223 0.2717172115005908
+5 0.4996775375595456 0.45831429696731 0.2141223891535361 0.14699094131547852
+3 0.2785489190179553 0.4670460811343048 0.15605716379626236 0.1442181961402127
+2 0.9004946866984243 0.5296691610870421 0.15558446317332356 0.10808979913351713
+2 0.1255661414437523 0.5302796376526192 0.1115243679003298 0.11231193383221742
+1 0.4928032246244045 0.5507640803465932 0.03995236350311469 0.03959826703426546
+1 0.42574935873946496 0.5712721543914927 0.06039941370465372 0.06589602205592753
+4 0.2794210333455478 0.6125797558093737 0.1554598754122389 0.1468964159117763
+1 0.14955294979846098 0.6205947223316266 0.05776474899230487 0.06033871602993304
+1 0.5632979113228289 0.6240055139818826 0.09681201905459876 0.11879086254430879
+4 0.9551264199340418 0.624671130366286 0.03138145840967387 0.03600630169358015
+4 0.09986075485525833 0.6260259944860181 0.0366397947966288 0.03662071681764474
+0 0.7255734701355808 0.6376841276092949 0.157248076218395 0.10410791650256006
+4 0.9577720776841334 0.6712997243009059 0.02884939538292415 0.04095313115399764
+4 0.12024184683034078 0.6721425758172509 0.07935507511909125 0.04052776683733754
+4 0.42443752290216197 0.6773060259944861 0.09104067423964822 0.03614415124064592
+4 0.8655734701355808 0.6776486805829066 0.08313301575668744 0.024363922804253643
+1 0.21567973616709418 0.1499369830641985 0.05060828142176622 0.06125640015754234
+4 0.782103334554782 0.154883812524616 0.060971051667277384 0.06163450177235132
+4 0.5012165628435324 0.1885191020086648 0.14333455478197143 0.03263489562820008
+4 0.9373506779039942 0.19026782197715636 0.05257237083180652 0.020949192595510042
+1 0.5020520337119824 0.2776092949980307 0.19535360938072555 0.11387554155179204
+1 0.12585928911689265 0.28114612051988974 0.10836203737632834 0.10729814887751084
+2 0.903081714913888 0.28357227254824735 0.15814950531330157 0.10031902323749509
+4 0.2814730670575302 0.3555612445844821 0.15768046903627703 0.07435604568727845

test_imgs/labels/03.txt

@@ -0,0 +1,19 @@
+4 0.21092948717948717 0.374340995053826 0.039067599067599074 0.014262438172825138
+3 0.21183275058275058 0.4295082921152168 0.0810576923076923 0.09298225196392201
+0 0.3031410256410257 0.47855106197265057 0.06031468531468531 0.1229735234215886
+4 0.2119871794871795 0.512737852778586 0.0850174825174825 0.07218795461157987
+4 0.4220775058275058 0.5193860925225487 0.013243006993006994 0.05630200756473669
+4 0.8641695804195804 0.534887983706721 0.0559965034965035 0.027390165842304335
+2 0.5031177156177157 0.5355600814663951 0.09374125874125874 0.03150130928135001
+4 0.5966083916083916 0.5386645330229852 0.08832459207459208 0.021905731742798955
+1 0.20175116550116548 0.18753564154786148 0.07063228438228439 0.05163514693046261
+2 0.8318851981351982 0.2160837940064009 0.04932983682983683 0.10572010474250801
+2 0.7959994172494173 0.21695955775385511 0.016526806526806527 0.06963922025021821
+2 0.7587004662004663 0.21790514983997675 0.019874708624708623 0.06936863543788187
+2 0.7202068764568764 0.23552807681117252 0.01865967365967366 0.10137038114634857
+4 0.7793181818181818 0.245010183299389 0.012392191142191143 0.10520221123072447
+4 0.7021124708624709 0.25278149549025314 0.01386946386946387 0.12506546406750074
+4 0.7421182983682983 0.2661798079720687 0.011608391608391609 0.14861798079720687
+4 0.5724475524475525 0.2722897876054699 0.014434731934731935 0.12386965376782078
+5 0.5188111888111888 0.28004364271166715 0.1105128205128205 0.26218795461157984
+4 0.25699009324009325 0.28994471923188825 0.1774562937062937 0.1543148094268257

test_imgs/labels/04.txt

@@ -0,0 +1,23 @@
+2 0.9087275449101796 0.40016541945647893 0.11264221556886228 0.1231744781410004
+3 0.2857784431137725 0.4036274123670736 0.1459693113772455 0.12429696730996455
+2 0.10357035928143712 0.4045844820795589 0.1486002994011976 0.1253564395431272
+5 0.5141504491017964 0.46024812918471836 0.21870134730538923 0.15662465537613235
+4 0.7409730538922156 0.5048011027963766 0.17294535928143714 0.38798345805435214
+4 0.27876497005988027 0.517093343836156 0.14128742514970058 0.10336746750689248
+2 0.9115980538922156 0.5259472233162663 0.10823727544910179 0.10866089011421819
+2 0.10225299401197606 0.5291492713666798 0.1424176646706587 0.11150059078377315
+1 0.5764408682634731 0.616423788893265 0.09887724550898203 0.11879086254430879
+1 0.1527432634730539 0.6205947223316266 0.058997005988023946 0.06033871602993304
+4 0.09265718562874252 0.6296455297361166 0.06111152694610778 0.03662071681764474
+0 0.2827919161676647 0.6440173296573455 0.15682634730538922 0.10556912170145727
+4 0.12400449101796407 0.6706183536825522 0.12190119760479043 0.03979125640015754
+4 0.43349176646706583 0.6773060259944861 0.09298278443113772 0.03614415124064592
+4 0.8840381736526947 0.6776486805829066 0.08490643712574851 0.024363922804253643
+1 0.23206586826347306 0.12258369436786136 0.07332335329341316 0.07564789287120914
+4 0.7973353293413173 0.12858605750295393 0.06583083832335329 0.021638440330838914
+4 0.9175935628742514 0.18396218983851909 0.0536938622754491 0.020949192595510042
+4 0.5119086826347305 0.1885191020086648 0.14639221556886228 0.03263489562820008
+1 0.5127619760479042 0.2776092949980307 0.19952095808383233 0.11387554155179204
+1 0.10677769461077845 0.27968491532099254 0.1556062874251497 0.10754233950374163
+2 0.9093937125748504 0.28222922410397794 0.11602919161676646 0.1006695549428909
+4 0.2884842814371258 0.3261835368255218 0.1565306886227545 0.027924379677038206

test_imgs/labels/05.txt

@@ -0,0 +1,27 @@
+2 0.9015390252839869 0.40270578968097676 0.15897764748992305 0.12198897203623474
+2 0.12520703554415535 0.4054430878298543 0.11080615610113594 0.11905474596297753
+4 0.7280505679736167 0.450106341079165 0.1641626969585929 0.27356833398975977
+5 0.4996775375595456 0.45831429696731 0.2141223891535361 0.14699094131547852
+3 0.27911322828875046 0.465959038991729 0.15640161231220226 0.1442181961402127
+2 0.9089739831440089 0.5284245766049626 0.13862953462806887 0.11057502953918866
+2 0.1255661414437523 0.5302796376526192 0.1115243679003298 0.11231193383221742
+1 0.4928032246244045 0.5507640803465932 0.03995236350311469 0.03959826703426546
+1 0.42574935873946496 0.5712721543914927 0.06039941370465372 0.06589602205592753
+4 0.27994869915720044 0.6077471445450965 0.15517039208501282 0.13965340685309177
+1 0.14955294979846098 0.6205947223316266 0.05776474899230487 0.06033871602993304
+1 0.5632979113228289 0.6240055139818826 0.09681201905459876 0.11879086254430879
+4 0.9551264199340418 0.624671130366286 0.03138145840967387 0.03600630169358015
+4 0.09986075485525833 0.6260259944860181 0.0366397947966288 0.03662071681764474
+0 0.7241993404177354 0.6386293816463174 0.15809820447050202 0.10563607719574634
+4 0.9577720776841334 0.6712997243009059 0.02884939538292415 0.04095313115399764
+4 0.12024184683034078 0.6721425758172509 0.07935507511909125 0.04052776683733754
+4 0.42443752290216197 0.6773060259944861 0.09104067423964822 0.03614415124064592
+4 0.8655734701355808 0.6776486805829066 0.08313301575668744 0.024363922804253643
+1 0.21567973616709418 0.1499369830641985 0.05060828142176622 0.06125640015754234
+4 0.782103334554782 0.154883812524616 0.060971051667277384 0.06163450177235132
+4 0.5012165628435324 0.1885191020086648 0.14333455478197143 0.03263489562820008
+4 0.9373506779039942 0.19026782197715636 0.05257237083180652 0.020949192595510042
+1 0.5020520337119824 0.2776092949980307 0.19535360938072555 0.11387554155179204
+1 0.1301209234151704 0.27997243009058687 0.10836203737632834 0.10729814887751084
+2 0.903081714913888 0.28357227254824735 0.15814950531330157 0.10031902323749509
+4 0.27973250274825945 0.35714060653800705 0.15549651887138147 0.07602599448601811

writeMD.py

@@ -0,0 +1,180 @@
+from mdutils.mdutils import MdUtils
+import os
+import cv2
+import numpy as np
+import requests
+
+
+# url = "http://192.168.199.249:4869"
+
+# 标签是经过归一化的，需要变回来
+# xywh格式 ---> box四个顶点的坐标
+def xywh2lrbt(img_w, img_h, box):
+    c, x, y, w, h = float(box[0]), float(box[1]), float(box[2]), float(box[3]), float(box[4])
+    x = x * img_w  # 中心坐标x
+    w = w * img_w  # box的宽
+    y = y * img_h  # 中心坐标y
+    h = h * img_h  # box的高
+
+    lt_x, lt_y = x - w / 2, y - h / 2  # left_top_x, left_top_y
+    lb_x, lb_y = x - w / 2, y + h / 2  # left_bottom_x, left_bottom_y
+    rb_x, rb_y = x + w / 2, y + h / 2  # right_bottom_x, right_bottom_y
+    rt_x, rt_y = x + w / 2, y - h / 2  # right_top_x, right_bottom_y
+
+    lrbt = [[lt_x, lt_y], [lb_x, lb_y], [rb_x, rb_y], [rt_x, rt_y]]
+
+    return lrbt, c
+
+
+def IOU(rect1, rect2):
+    xmin1, ymin1, xmax1, ymax1 = rect1
+    xmin2, ymin2, xmax2, ymax2 = rect2
+    s1 = (xmax1 - xmin1) * (ymax1 - ymin1)
+    s2 = (xmax2 - xmin2) * (ymax2 - ymin2)
+
+    sum_area = s1 + s2
+
+    left = max(xmin2, xmin1)
+    right = min(xmax2, xmax1)
+    top = max(ymin2, ymin1)
+    bottom = min(ymax2, ymax1)
+
+    if left >= right or top >= bottom:
+        return 0
+
+    intersection = (right - left) * (bottom - top)
+    return intersection / (sum_area - intersection) * 1.0
+
+
+def send_requests():
+    return result
+
+img_path = "./images/test1.jpg"
+b_img_list = []
+b_img_list.append(('file_list', ('image.jpeg', open(img_path, 'rb'), 'image/jpeg')))
+payload = {'model_name': 'ocr-layout', 'img_size': 800, 'download_image': False}
+response = requests.request("POST", "http://192.168.199.107:18089/detect", headers={}, data=payload, files=b_img_list)
+result = eval(str(response.text))
+result = eval(str(result['result']))
+# print(len(result))
+
+mdFile = MdUtils(file_name='yolov5_yili_0818_1', title='Markdown File Example')
+mdFile.new_header(level=1, title="yolov5预测结果")
+list_item = ["ori_cla", "predict_cla", "cut_ori_img", "cut_predict_img", "isTrue"]
+# list_item.extend(["ori_cla", "predict_cla", "cut_ori_img", "cut_predict_img", "isTrue"])
+# for i in range(len(result[0])):
+#     # print(stencil_list[0][i])
+#
+#     predict_cla = result[0][i]["class"]
+
+# 获取真实的标签
+label = "./images/test1.txt"
+with open(label, mode="r", encoding="utf-8") as f:
+    lines = f.readlines()
+
+img = cv2.imread(img_path)
+img_h, img_w, _ = img.shape
+
+arr_label = []  # labels的边界框，存储四个顶点坐标
+ori_clas = []  # label的真实类别
+for line in lines:
+    line = line.split(" ")
+    lrbt, c = xywh2lrbt(img_w, img_h, line)
+    arr_label.append(lrbt)
+    ori_clas.append(c)
+
+# cut_imgs = []  # 被剪切的图像
+num = 0
+for i in range(len(arr_label)):
+    lt_x, lt_y = int(arr_label[i][0][0]), int(arr_label[i][0][1])
+    rb_x, rb_y = int(arr_label[i][2][0]), int(arr_label[i][2][1])
+    rect1 = [lt_x, lt_y, rb_x, rb_y]
+
+    cut_ori_img = img[lt_y:rb_y, lt_x:rb_x, :]
+    save_cut_ori_img = "./images/cut_imgs/" + str(i) + "_" + str(num) + ".jpg"
+    cv2.imwrite(save_cut_ori_img, cut_ori_img)
+    save_cut_ori_img = mdFile.new_inline_image(text='', path=save_cut_ori_img)
+    num += 1
+    ori_cla = ori_clas[i]
+    for j in range(len(result[0])):
+        res = []
+        predict_cla = result[0][j]['class']
+        if ori_cla == predict_cla:
+            rect2 = result[0][j]['bbox']
+            iou = IOU(rect1, rect2)
+            if iou >= 0.5:
+                cut_predict_img = img[rect2[1]:rect2[3], rect2[0]:rect2[2], :]
+                save_cut_predict_img = "./images/cut_imgs/" + str(j) + ".jpg"
+                cv2.imwrite(save_cut_predict_img, cut_predict_img)
+                save_cut_predict_img = mdFile.new_inline_image(text='', path=save_cut_predict_img)
+                isTrue = True
+                list_item.extend([ori_cla, predict_cla, save_cut_ori_img, save_cut_predict_img, isTrue])
+                # list_item.extend(res)
+                break
+        elif j == len(result[0]) - 1:
+            isTrue = False
+            list_item.extend([ori_cla, predict_cla, save_cut_ori_img, "", isTrue])
+            # res.append([ori_cla, predict_cla, "", "", isTrue])
+            list_item.extend(res)
+#
+# print(list_item)
+# print(len(arr_label))
+print(len(list_item))
+# print(len(result[0]))
+mdFile.new_line()
+mdFile.new_table(columns=5, rows=len(list_item)//5, text=list_item, text_align='center')
+mdFile.create_md_file()
+
+if __name__ == "__main__":
+    images_path = "./test_imgs/images/"
+    labels_path = "./test_imgs/labels/"
+    list_images = os.listdir(images_path)
+    list_labels = os.listdir(labels_path)
+    for image in list_images:
+        img_name = image.split(".")[0]
+        mdFile = MdUtils(file_name='yq_0819_' + str(img_name), title='yq_0819_' + str(img_name) + '_iou>=0.5')
+
+
+
+
+#
+# images = os.listdir(images_path)
+# labels = os.listdir(labels_path)
+#
+# assert len(images) == len(labels)
+#
+# for i in range(len(images)):
+#     image = images[i]
+#     label = labels[i]
+#     with open(labels_path + label, mode="r", encoding="utf-8") as f:
+#         lines = f.readlines()
+#
+#     img = images_path + images[i]
+#     img = cv2.imread(img)
+#     img_h, img_w, _ = img.shape
+#
+#     arr_label = []  # labels的边界框，存储四个顶点坐标
+#     ori_cla = []  # label的真实类别
+#     for line in lines:
+#         line = line.split(" ")
+#         lrbt, c = xywh2lrbt(img_w, img_h, line)
+#         arr_label.append(lrbt)
+#         ori_cla.append(c)
+#
+#     cut_path = "./valid_img/cut/"
+#
+#     for i in range(len(arr_label)):
+#         lt_x, lt_y = int(arr_label[i][0][0]), int(arr_label[i][0][1])
+#         rb_x, rb_y = int(arr_label[i][2][0]), int(arr_label[i][2][1])
+#         cut_img = img[lt_y:rb_y, lt_x:rb_x, :]
+#         cv2.imwrite(cut_path + str(i) + ".jpg", cut_img)
+#
+#         box = np.reshape(np.array(arr_label[i]), [-1, 1, 2]).astype(np.int32)
+#         img = cv2.polylines(np.array(img), [box], True, (255, 0, 0), 5)
+#
+#
+#
+#
+# mdFile.create_md_file()
+#
+#