Differences

This shows you the differences between two versions of the page.

--- projects:plate [2022/05/08 22:35] – old revision restored (2022/02/25 12:05) 216.244.66.228
+++ projects:plate [2022/06/20 14:19] (current) – old revision restored (2022/03/12 21:34) 154.54.249.201
@@ Line 1: / Line 1: @@
-====== plate ======
+====== yolo train ======
-plate detection with neural network
+<code>
-  * https://matthewearl.github.io/2016/05/06/cnn-anpr/
+git clone https://github.com/puzzledqs/BBox-Label-Tool.git
-  * https://github.com/matthewearl/deep-anpr
+</code>
-http://www.pyimagesearch.com/2017/02/13/recognizing-digits-with-opencv-and-python/
+<file python convert.py>
-<code bash>
+import os
-# load the example image
+from os import walk, getcwd
-image = cv2.imread("example.jpg")
+from PIL import Image
-# pre-process the image by resizing it, converting it to
-# graycale, blurring it, and computing an edge map
-image = imutils.resize(image, height=500)
-gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
-# Applying Gaussian blurring with a 5×5 kernel to reduce high-frequency noise
-blurred = cv2.GaussianBlur(gray, (5, 5), 0)
-# Computing the edge map via the Canny edge detector.
+classes = ["targa"]
-edged = cv2.Canny(blurred, 50, 200, 255)
-# find contours in the edge map, then sort them by their
+def convert(size, box):
-# size in descending order
+    dw = 1./size[0]
-cnts = cv2.findContours(edged.copy(), cv2.RETR_EXTERNAL,
+    dh = 1./size[1]
-	cv2.CHAIN_APPROX_SIMPLE)
+    x = (box[0] + box[1])/2.0
-cnts = cnts[0] if imutils.is_cv2() else cnts[1]
+    y = (box[2] + box[3])/2.0
-cnts = sorted(cnts, key=cv2.contourArea, reverse=True)
+    w = box[1] - box[0]
-displayCnt = None
+    h = box[3] - box[2]
+    x = x*dw
-# loop over the contours
+    w = w*dw
-for c in cnts:
+    y = y*dh
-	# approximate the contour
+    h = h*dh
-	peri = cv2.arcLength(c, True)
+    return (x,y,w,h)
-	approx = cv2.approxPolyDP(c, 0.02 * peri, True)
-	# if the contour has four vertices, then we have found
+"""-------------------------------------------------------------------"""
-	# the thermostat display
-	if len(approx) == 4:
-		displayCnt = approx
-		break
-# extract the plate, apply a perspective transform to it
+""" Configure Paths"""
-# Applying this perspective transform gives us a top-down, birds-eye-view of plate
+mypath = "Labels/001/"
-warped = four_point_transform(gray, displayCnt.reshape(4, 2))
+outpath = "Labels/output/"
-output = four_point_transform(image, displayCnt.reshape(4, 2))
-# threshold the warped image, then apply a series of morphological
-# operations to cleanup the thresholded image
-thresh = cv2.threshold(warped, 0, 255,
-	cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU)[1]
-kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (1, 5))
-thresh = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, kernel)
-# find contours in the thresholded image, then initialize the
+cls = "001"
-# digit contours lists
-cnts = cv2.findContours(thresh.copy(), cv2.RETR_EXTERNAL,
-	cv2.CHAIN_APPROX_SIMPLE)
-cnts = cnts[0] if imutils.is_cv2() else cnts[1]
-digitCnts = []
-# loop over the digit area candidates
-for c in cnts:
-	# compute the bounding box of the contour
-	(x, y, w, h) = cv2.boundingRect(c)
-	# if the contour is sufficiently large, it must be a digit
-	if w >= 15 and (h >= 30 and h <= 40):
-		digitCnts.append(c)
-# TODO display contour
-# cv2.rectangle(output, (x, y), (x + w, y + h), (0, 255, 0), 1)
+wd = getcwd()
+list_file = open('%s/%s_list.txt'%(wd, cls), 'w')
-# sort the contours from left-to-right, then initialize the
+""" Get input text file list """
-# actual digits themselves
+txt_name_list = []
-digitCnts = contours.sort_contours(digitCnts,
+for (dirpath, dirnames, filenames) in walk(mypath):
-	method="left-to-right")[0]
+    print(filenames)
-digits = []
+    txt_name_list.extend(filenames)
+    break
+print(txt_name_list)
-# 	cv2.putText(output, str(digit), (x - 10, y - 10),
+""" Process """
-#		cv2.FONT_HERSHEY_SIMPLEX, 0.65, (0, 255, 0), 2)
+for txt_name in txt_name_list:
+    # txt_file =  open("Labels/stop_sign/001.txt", "r")
+    """ Open input text files """
+    txt_path = mypath + txt_name
+    print("Input:" + txt_path)
+    txt_file = open(txt_path, "r")
+    lines = txt_file.read().split('\r\n')   #for ubuntu, use "\r\n" instead of "\n"
+    """ Open output text files """
+    txt_outpath = outpath + txt_name
+    print("Output:" + txt_outpath)
+    txt_outfile = open(txt_outpath, "w")
+    """ Convert the data to YOLO format """
+    ct = 0
+    for line in lines:
+        #print('lenth of line is: ')
+        #print(len(line))
+        #print('\n')
+        if(len(line) >= 2):
+            ct = ct + 1
+            print(line + "\n")
+            elems = line.split(' ')
+            print(elems)
+            cls_id = elems[0].split('\n')[0]
+            xmin = elems[0].split('\n')[1]
+            xmax = elems[2]
+            ymin = elems[1]
+            ymax = elems[3][:-1]
+            #
+            img_path = str('%s/Images/%s/%s.JPEG'%(wd, cls, os.path.splitext(txt_name)[0]))
+            #t = magic.from_file(img_path)
+            #wh= re.search('(\d+) x (\d+)', t).groups()
+            im=Image.open(img_path)
+            w= int(im.size[0])
+            h= int(im.size[1])
+            #w = int(xmax) - int(xmin)
+            #h = int(ymax) - int(ymin)
+            # print(xmin)
+            print(w, h)
+            b = (float(xmin), float(xmax), float(ymin), float(ymax))
+            bb = convert((w,h), b)
+            print(bb)
+            txt_outfile.write(str(cls_id) + " " + " ".join([str(a) for a in bb]) + '\n')
+    """ Save those images with bb into list"""
+    if(ct != 0):
+        list_file.write('%s/images/%s/%s.JPEG\n'%(wd, cls, os.path.splitext(txt_name)[0]))
+list_file.close()
+</file>
+Train.txt Text.txt
+<file python process.py>
+import glob, os
+# Current directory
+current_dir = os.path.dirname(os.path.abspath(__file__))
+# Directory where the data will reside, relative to 'darknet.exe'
+path_data = '*IMAGE DIRECTORY*'
+# Percentage of images to be used for the test set
+percentage_test = 10;
+# Create and/or truncate train.txt and test.txt
+file_train = open('train.txt', 'w')
+file_test = open('test.txt', 'w')
+# Populate train.txt and test.txt
+counter = 1
+index_test = round(100 / percentage_test)
+for pathAndFilename in glob.iglob(os.path.join(current_dir, "*.JPEG")):
+    title, ext = os.path.splitext(os.path.basename(pathAndFilename))
+    if counter == index_test:
+        counter = 1
+        file_test.write(path_data + title + '.JPEG' + "\n")
+    else:
+        file_train.write(path_data + title + '.JPEG' + "\n")
+        counter = counter + 1
+</file>
+Put images inside BBox-Label-Tool/Images/001/
+convert to JPEG and delete old images
+<code>
+mogrify -format JPEG *.jpg
+rm *.jpg
 </code>
+Go to main folder and run python main.py
+<code>
+python main.py
+</code>
+Write 001 inside Image Dir box and load
+Create a label for each image
+After that, exit and create a new directory inside Label
+<code>
+mkdir output
+</code>
+Run convert.py
+<code>
+python convert.py
+</code>
+Now create test.txt and train.txt with process.py
+<code>
+python process.py
+</code>
+<code>
+├── Images (input)
+│   ├── 001
+│   │   ├── 20180319_113309.JPEG
+│   └── targa
+├── Labels (output)
+│   ├── 001
+│   │   ├── 20180319_113309.txt
+│   └── output
+│       ├── 20180319_113309.txt
+</code>
+====== Darknet ======
+<code>
+git clone https://github.com/pjreddie/darknet
+cd darknet
+make
+</code>
+Copy train.txt and test.txt inside darknet/cfg/
+Create 3 files:
+obj.data
+obj.names
+obj.cfg
+<file obj.data>
+classes= *NUMBER CLASSES*
+train  = *TRAIN DIRECTORY+
+valid = *TEST DIRECTORY*
+names = obj.names
+backup = *BACKUP FOLDER*
+</file>
+<file obj.names>
+*CLASS NAME*
+</file>
+Copy yolov2-tiny.cfg and change [region]:classes to
+classes = *NUMBER CLASSES*
+filters = (*NUMBER CLASSES* +5)*5