优化代码，修复一些bug，新增python版本nms

README.md

@@ -7,6 +7,7 @@
 
 
 ## 环境部署
+
 GPU部署 参考:setup.md     
 GPU部署 参考:setup-cpu.md   
 

apphelper/image.py

@@ -18,6 +18,10 @@
 
 def sort_box_(box):
     x1,y1,x2,y2,x3,y3,x4,y4 = box[:8]
+    pts = (x1,y1),(x2,y2),(x3,y3),(x4,y4)
+    pts = np.array(pts, dtype="float32")
+    (x1,y1),(x2,y2),(x3,y3),(x4,y4) = _order_points(pts)
+    """
     newBox = [[x1,y1],[x2,y2],[x3,y3],[x4,y4]]
     ## sort x
     newBox = sorted(newBox,key=lambda x:x[0])
@@ -35,8 +39,42 @@ def sort_box_(box):
     
     newBox = sorted(newBox,key=lambda x:-x[1])
     x3,y3 = sorted(newBox[:2],key=lambda x:x[0])[0]
+    """
     return x1,y1,x2,y2,x3,y3,x4,y4
 
+
+import numpy as np
+from scipy.spatial import distance as dist
+def _order_points(pts):
+    # 根据x坐标对点进行排序
+    """
+    --------------------- 
+    作者：Tong_T 
+    来源：CSDN 
+    原文：https://blog.csdn.net/Tong_T/article/details/81907132 
+    版权声明：本文为博主原创文章，转载请附上博文链接！
+    """
+    x_sorted = pts[np.argsort(pts[:, 0]), :]
+
+    # 从排序中获取最左侧和最右侧的点
+    # x坐标点
+    left_most = x_sorted[:2, :]
+    right_most = x_sorted[2:, :]
+
+    # 现在，根据它们的y坐标对最左边的坐标进行排序，这样我们就可以分别抓住左上角和左下角
+    left_most = left_most[np.argsort(left_most[:, 1]), :]
+    (tl, bl) = left_most
+
+    # 现在我们有了左上角坐标，用它作为锚来计算左上角和右上角之间的欧氏距离;
+    # 根据毕达哥拉斯定理，距离最大的点将是我们的右下角
+    distance = dist.cdist(tl[np.newaxis], right_most, "euclidean")[0]
+    (br, tr) = right_most[np.argsort(distance)[::-1], :]
+
+    # 返回左上角，右上角，右下角和左下角的坐标
+    return np.array([tl, tr, br, bl], dtype="float32")
+
+
+
 def solve(box):
      """
      绕 cx,cy点 w,h 旋转 angle 的坐标
@@ -96,10 +134,18 @@ def read_voc_xml(p):
                     continue
 
                 angle = np.float(angle)
+
+                if abs(angle)>np.pi/2:
+                    w,h = h,w
+                    angle     = abs(angle)%(np.pi/2)*np.sign(angle)
+
                 x1,y1,x2,y2,x3,y3,x4,y4 = xy_rotate_box(cx,cy,w,h,angle)
-                if angle>np.pi/4:
+                x1,y1,x2,y2,x3,y3,x4,y4 = sort_box_([x1,y1,x2,y2,x3,y3,x4,y4])
+                """
+                if abs(angle)>np.pi/2:
                     ##lableImg bug
                     x1,y1,x2,y2,x3,y3,x4,y4 = sort_box_([x1,y1,x2,y2,x3,y3,x4,y4])
+                """
                 angle,w,h,cx,cy = solve([x1,y1,x2,y2,x3,y3,x4,y4])
 
             else:

application/README.md

@@ -0,0 +1 @@
+基于该项目实现财务票据、表格文字的识别（将于近期公布）

config.py

@@ -1,31 +1,53 @@
 import os
-opencvFlag = 'keras'##keras,opencv,darknet
+########################文字检测########################
+##文字检测引擎 keras,opencv,darknet
+pwd = os.getcwd()
+opencvFlag = 'keras'
+IMGSIZE = (608,608)## yolo3 输入图像尺寸
+## keras 版本anchors
 keras_anchors = '8,9, 8,18, 8,31, 8,59, 8,124, 8,351, 8,509, 8,605, 8,800'
 class_names = ['none','text',]
-GPU = True##OCR 是否启用GPU
-GPUID=0##调用GPU序号
+kerasTextModel=os.path.join(pwd,"models","text.h5")##keras版本模型权重文件
 
+############## darknet yolo  ##############
 darknetRoot = os.path.join(os.path.curdir,"darknet")## yolo 安装目录
-pwd = os.getcwd()
-yoloCfg = os.path.join(pwd,"models","text.cfg")
+yoloCfg     = os.path.join(pwd,"models","text.cfg")
 yoloWeights = os.path.join(pwd,"models","text.weights")
-yoloData = os.path.join(pwd,"models","text.data")
+yoloData    = os.path.join(pwd,"models","text.data")
+############## darknet yolo  ##############
+
+########################文字检测########################
+
+## GPU选择及启动GPU序号
+GPU = True##OCR 是否启用GPU
+GPUID=0##调用GPU序号
 
-kerasTextModel=os.path.join(pwd,"models","text.h5")##keras版本模型
-##文字方向检测
+## nms选择,支持cython,gpu,python
+nmsFlag='gpu'## cython/gpu/python
+
+
+
+##vgg文字方向检测模型
+DETECTANGLE=True##是否进行文字方向检测
 AngleModelPb = os.path.join(pwd,"models","Angle-model.pb")
 AngleModelPbtxt = os.path.join(pwd,"models","Angle-model.pbtxt")
-IMGSIZE = (608,608)## yolo3 输入图像尺寸
+
+
+######################OCR模型######################
 ##是否启用LSTM crnn模型
-DETECTANGLE=True##是否进行文字方向检测
-LSTMFLAG = True##OCR模型是否调用LSTM层
-chinsesModel = True##模型选择 True:中英文模型 False:英文模型
+##OCR模型是否调用LSTM层
+LSTMFLAG = True
+##模型选择 True:中英文模型 False:英文模型
+
+chinsesModel = True
 
 if chinsesModel:
     if LSTMFLAG:
         ocrModel  = os.path.join(pwd,"models","ocr-lstm.pth")
     else:
         ocrModel = os.path.join(pwd,"models","ocr-dense.pth")
 else:
+        ##纯英文模型
         LSTMFLAG=True
         ocrModel = os.path.join(pwd,"models","ocr-english.pth")
+######################OCR模型######################

crnn/crnn.py

@@ -1,38 +1,36 @@
 #coding:utf-8
-import sys
-sys.path.insert(1, "./crnn")
 import torch
 import torch.utils.data
 from torch.autograd import Variable 
 from crnn import util
 from crnn import dataset
-from crnn.models import crnn as crnn
+from crnn.network import CRNN
 from crnn import keys
 from collections import OrderedDict
 from config import ocrModel,LSTMFLAG,GPU
 from config import chinsesModel
 def crnnSource():
     if chinsesModel:
-        alphabet = keys.alphabetChinese
+        alphabet = keys.alphabetChinese##中英文模型
     else:
-        alphabet = keys.alphabetEnglish
+        alphabet = keys.alphabetEnglish##英文模型
 
     converter = util.strLabelConverter(alphabet)
     if torch.cuda.is_available() and GPU:
-        model = crnn.CRNN(32, 1, len(alphabet)+1, 256, 1,lstmFlag=LSTMFLAG).cuda()##LSTMFLAG=True crnn 否则 dense ocr
+        model = CRNN(32, 1, len(alphabet)+1, 256, 1,lstmFlag=LSTMFLAG).cuda()##LSTMFLAG=True crnn 否则 dense ocr
     else:
-        model = crnn.CRNN(32, 1, len(alphabet)+1, 256, 1,lstmFlag=LSTMFLAG).cpu()
+        model = CRNN(32, 1, len(alphabet)+1, 256, 1,lstmFlag=LSTMFLAG).cpu()
 
-    state_dict = torch.load(ocrModel,map_location=lambda storage, loc: storage)
-    new_state_dict = OrderedDict()
-    for k, v in state_dict.items():
+    trainWeights = torch.load(ocrModel,map_location=lambda storage, loc: storage)
+    modelWeights = OrderedDict()
+    for k, v in trainWeights.items():
         name = k.replace('module.','') # remove `module.`
-        new_state_dict[name] = v
+        modelWeights[name] = v
     # load params
 
-    model.load_state_dict(new_state_dict)
+    model.load_state_dict(modelWeights)
     model.eval()
-    
+
     return model,converter
 
 ##加载模型
@@ -41,30 +39,25 @@ def crnnSource():
 def crnnOcr(image):
        """
        crnn模型，ocr识别
-       @@model,
-       @@converter,
-       @@im
-       @@text_recs:text box
-
+       image:PIL.Image.convert("L")
        """
        scale = image.size[1]*1.0 / 32
        w = image.size[0] / scale
        w = int(w)
-       #print "im size:{},{}".format(image.size,w)
        transformer = dataset.resizeNormalize((w, 32))
        if torch.cuda.is_available() and GPU:
-           image = transformer(image).cuda()
+           image   = transformer(image).cuda()
        else:
-           image = transformer(image).cpu()
+           image   = transformer(image).cpu()
 
-       image = image.view(1, *image.size())
-       image = Variable(image)
+       image       = image.view(1, *image.size())
+       image       = Variable(image)
        model.eval()
-       preds = model(image)
-       _, preds = preds.max(2)
-       preds = preds.transpose(1, 0).contiguous().view(-1)
-       preds_size = Variable(torch.IntTensor([preds.size(0)]))
-       sim_pred = converter.decode(preds.data, preds_size.data, raw=False)
+       preds       = model(image)
+       _, preds    = preds.max(2)
+       preds       = preds.transpose(1, 0).contiguous().view(-1)
+       preds_size  = Variable(torch.IntTensor([preds.size(0)]))
+       sim_pred    = converter.decode(preds.data, preds_size.data, raw=False)
 
        return sim_pred
 

model.py

@@ -18,7 +18,8 @@
 else:
     ## keras版本文字检测
     from text import keras_detect as detect
-
+
+print("Text detect engine:{}".format(opencvFlag))
 
 def text_detect(img,
                 MAX_HORIZONTAL_GAP=30,
@@ -34,8 +35,6 @@ def text_detect(img,
 
                 ):
     boxes, scores = detect.text_detect(np.array(img))
-
-
     boxes = np.array(boxes,dtype=np.float32)
     scores = np.array(scores,dtype=np.float32)
     textdetector  = TextDetector(MAX_HORIZONTAL_GAP,MIN_V_OVERLAPS,MIN_SIZE_SIM)

setup-cpu.md

@@ -11,5 +11,6 @@ pip install keras==2.1.5 tensorflow==1.8
 conda install pytorch torchvision -c pytorch
 ## linux
 ## conda install pytorch-cpu torchvision-cpu -c pytorch
+## python版本nms无须执行下一步   
 pushd text/detector/utils && sh make-for-cpu.sh && popd
 

setup.md

@@ -7,7 +7,8 @@ pip install -U pillow -i https://pypi.tuna.tsinghua.edu.cn/simple/
 pip install keras==2.1.5 tensorflow==1.8 tensorflow-gpu==1.8      
 pip install web.py==0.40.dev0       
 conda install pytorch torchvision -c pytorch          
-## pip install torch torchvision      
+## pip install torch torchvision   
+## python版本nms无须执行下一步      
 pushd text/detector/utils && sh make.sh && popd      
 
 

text/detector/detectors.py

@@ -1,27 +1,49 @@
 #coding:utf-8
 import numpy as np
-from config import GPUID,GPU
-from text.detector.utils.cython_nms import nms as cython_nms
+from config import GPUID,GPU,nmsFlag
+from text.detector.utils.python_nms import nms as python_nms ##python版本nms
+
 from text.detector.text_proposal_connector import TextProposalConnector
 
-##优先加载编译对GPU编译的gpu_nms 如果不想调用GPU，在程序启动执行os.environ["CUDA_VISIBLE_DEVICES"] = "0"
-if GPU:
+
+##优先加载编译对GPU编译的gpu_nms
+
+if nmsFlag=='gpu' and GPU and GPUID is not None:
+    try:
+        from text.detector.utils.gpu_nms import gpu_nms
+    except:
+            gpu_nms = None
+    cython_nms = None
+
+elif nmsFlag=='python':
+     gpu_nms ==None
+     cython_nms = None
+
+elif nmsFlag=='cython':
     try:
-        from detector.utils.gpu_nms import gpu_nms
+        from text.detector.utils.cython_nms  import nms as  cython_nms
     except:
-        gpu_nms =cython_nms
+         cython_nms = None
+    gpu_nms ==None
+else:
+    gpu_nms =None
+    cython_nms = None
+
+print("Nms engine gpu_nms:",gpu_nms,",cython_nms:",cython_nms,",python_nms:",python_nms)
+
 
 def nms(dets, thresh):
     if dets.shape[0] == 0:
         return []
 
-    try:
-        if GPU and GPUID is not None:
-            return gpu_nms(dets, thresh, device_id=GPUID)
-    except:
-        pass
+    if  gpu_nms is not None and GPUID is not None:
+        return gpu_nms(dets, thresh, device_id=GPUID)
+
+    elif  cython_nms is not None:
+          return cython_nms(dets, thresh)
+    else:
+          return python_nms(dets, thresh, method='Union')
 
-    return cython_nms(dets, thresh)
 
 def normalize(data):
     if data.shape[0]==0:

text/detector/utils/__init__.py

@@ -1,6 +1 @@
-from . import cython_nms
-try:
-  from . import gpu_nms
-except:
-   gpu_nms = cython_nms
-
+