First working version

First working version of component Detector with ipython notebook example included

.gitattributes

@@ -1,2 +1,3 @@
 README.md filter=lfs diff=lfs merge=lfs -text
 __init__.py filter=lfs diff=lfs merge=lfs -text
+*.caffemodel filter=lfs diff=lfs merge=lfs -text

MANIFEST.in

@@ -1 +1 @@
-recursive-include detector/models
+recursive-include detector/models *.*

detector/componentDetector.py

@@ -8,7 +8,7 @@
 from PIL import Image, ImageDraw, ImageFont
 import argparse
 import cv2
-import copy, time
+import copy, time, math
 from . import config
 
 sys.path.insert(0, os.path.join(config.caffe_root, 'python'))
@@ -53,10 +53,11 @@ def detect(self, image, minComponentSize):
                 # sliding window
                 total_boxes = []
 
+                print("Computing bounding boxes")
                 start1 = time.time()
                 for (x,y,imw) in sliding_window(ims, (515, 515), (minComponentSize, minComponentSize)):
                     # pass the image window to the classifier network
-                    out = net_full_conv.forward_all(data=np.asarray([transformer.preprocess('data', imw)]))
+                    out = self.net_full_conv.forward_all(data=np.asarray([self.transformer.preprocess('data', imw)]))
 
                     # get the probability matrix associated with class 1 (IC here)
                     outprob = out['prob'][0,1]
@@ -90,15 +91,17 @@ def detect(self, image, minComponentSize):
                 end1 = time.time()
                 #print "Total time take for this image: " % (end1-start1)
 
+                print("Performing Non-Maxima Suppression")
                 # perform nms for the entire set of boxes
                 boxes_nms = np.array(total_boxes)
                 true_boxes = nms_max(boxes_nms, overlapThresh=0.2)
 
-
+                print("Clustering bounding boxes")
                 # Cluster the overlapping bounding boxes
                 clusters = clusterBoxes(true_boxes)
                 fCluster = getAvgClusterBoxes(clusters)
 
+                print("Finding enclosing boxes")
                 # group enclosed boxes
                 finalBoxes = enclosingBoxes(fCluster)
 

detector/helperModules.py

@@ -246,8 +246,8 @@ def loadNet(deployProto, fullConvProto, caffeModel, mode='CPU'):
 
     conv_params = {pr: (net_full_conv.params[pr][0].data, net_full_conv.params[pr][1].data) for pr in params_full_conv}
 
-    for conv in params_full_conv:
-        print '{} weights are {} dimensional and biases are {} dimensional'.format(conv, conv_params[conv][0].shape, conv_params[conv][1].shape)
+    #for conv in params_full_conv:
+    #    print '{} weights are {} dimensional and biases are {} dimensional'.format(conv, conv_params[conv][0].shape, conv_params[conv][1].shape)
 
     for pr, pr_conv in zip(cfg.params, params_full_conv):
         conv_params[pr_conv][0].flat = fc_params[pr][0].flat # flat unrolls the arrays

detector/nms.py

@@ -5,9 +5,9 @@
 
 
 def match_score(b1, b2):
-    # sum of the distance between centers of the two bounding 
+    # sum of the distance between centers of the two bounding
     # boxes and the intersection area of the boxes
-    
+
     # calculate the centers
     xc1 = (b1[0]+b1[2]) / 2
     yc1 = (b1[1]+b1[3]) / 2
@@ -22,22 +22,22 @@ def match_score(b1, b2):
         area = dx*dy
     else:
         area = 0
-    
+
     return dist+area
 
 #def box_merge(b1, b2):
-       
+
 
 def combine_boxes(boxes, overlapThresh=0.07):
     # Combines boxes across scales
     if len(boxes) == 0:
         return []
-    
-    # Bs - set of bounding boxes predicted by the regressor network 
+
+    # Bs - set of bounding boxes predicted by the regressor network
     # for each class in Cs across all spatial locations at scale s
     # In our case it the boxes variable itself
 
-    
+
 
 def nms_max(boxes, overlapThresh=0.3):
     if len(boxes) == 0:
@@ -74,15 +74,15 @@ def nms_max(boxes, overlapThresh=0.3):
         # compute the width and height of the bounding box
         w = np.maximum(0, xx2 - xx1 + 1)
         h = np.maximum(0, yy2 - yy1 + 1)
-        
+
         # area of i.
         area_i = np.maximum(0, x2[i] - x1[i] + 1) * np.maximum(0, y2[i] - y1[i] + 1)
         area_array = np.zeros(len(idxs) - 1)
         area_array.fill(area_i)
-        
+
         # compute the ratio of overlap
         overlap = (w * h) / (area[idxs[:last]]  - w * h + area_array)
-        
+
         # delete all indexes from the index list that have
         idxs = np.delete(idxs, np.concatenate(([last],np.where(overlap > overlapThresh)[0])))
 
@@ -94,7 +94,7 @@ def nms_average(boxes, overlapThresh=0.2):
     result_boxes = []
     if len(boxes) == 0:
         return []
-    
+
     # initialize the list of picked indexes
     pick = []
 
@@ -117,7 +117,7 @@ def nms_average(boxes, overlapThresh=0.2):
         i = idxs[last]
         pick.append(i)
 
-        # find the largest (x, y) coordinates for the start of the bounding box 
+        # find the largest (x, y) coordinates for the start of the bounding box
         # and the smallest (x, y) coordinates for the end of the bounding box
         xx1 = np.maximum(x1[i], x1[idxs[:last]])
         yy1 = np.maximum(y1[i], y1[idxs[:last]])
@@ -127,15 +127,15 @@ def nms_average(boxes, overlapThresh=0.2):
         # compute the width and height of the bounding box
         w = np.maximum(0, xx2 - xx1 + 1)
         h = np.maximum(0, yy2 - yy1 + 1)
-        
+
         # area of i.
         area_i = np.maximum(0, x2[i] - x1[i] + 1) * np.maximum(0, y2[i] - y1[i] + 1)
         area_array = np.zeros(len(idxs) - 1)
         area_array.fill(area_i)
 
         # compute the ratio of overlap
         overlap = (w * h) / (area[idxs[:last]])
-        
+
         delete_idxs = np.concatenate(([last],np.where(overlap > overlapThresh)[0]))
         # print "delete_idxs:", delete_idxs
         xmin = 10000
@@ -165,12 +165,11 @@ def nms_average(boxes, overlapThresh=0.2):
             xmax = x2[i]  + 0.1 * width
         if( ymax - y2[i] > 0.1 * height):
             ymax = y2[i] + 0.1 * height
-        
+
         result_boxes.append([xmin, ymin, xmax, ymax, ave_prob / len(delete_idxs)])
-	
+
         # delete all indexes from the index list that have
         idxs = np.delete(idxs, delete_idxs)
 
     # return only the bounding boxes that were picked using the integer data type
     return result_boxes
-

detector/windowExtractor.py

@@ -14,15 +14,15 @@
 def getWindowDims(image, winSize=[cfg.winSize, cfg.winSize], stride=[cfg.stride,cfg.stride]):
     # This function returns the bounding boxes for each window in the given image
     windows = []
-    
+
     for y in xrange(0, image.shape[0]-winSize[0], stride[0]):
         for x in xrange(0, image.shape[1]-winSize[1], stride[1]):
             windows.append([x, y, winSize[0], winSize[1]]) # row, col, width, height
 
     return windows
 
 def extractSubWindows(image, windows, transformer):
-    # This function extracts the subwindows and puts it in the format required by caffe for batch processing. 
+    # This function extracts the subwindows and puts it in the format required by caffe for batch processing.
     # Each window is of the same size
 
     subWindows = np.zeros(np.array([len(windows), 3, windows[0][2], windows[0][3]]))
@@ -45,4 +45,3 @@ def extractSubImages(image, windows):
         subWindows[i] = image[y:y+h, x:x+w]
 
     return subWindows
-

examples/IC_Detector.ipynb

@@ -13,48 +13,58 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
-   "metadata": {
-    "collapsed": true
-   },
+   "execution_count": 1,
+   "metadata": {},
    "outputs": [],
    "source": [
     "%matplotlib inline\n",
     "import sys\n",
     "import os\n",
     "import matplotlib.pyplot as plt\n",
     "import cv2\n",
-    "import config\n",
     "import caffe\n",
     "\n",
-    "from detector import componenetDetector as cd"
+    "from detector import componentDetector as cd\n",
+    "from detector import helperModules"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": null,
-   "metadata": {
-    "collapsed": true
-   },
-   "outputs": [],
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "[['images/A029F090.jpg', '59']]\n"
+     ]
+    }
+   ],
    "source": [
     "# load the image list file - format is full_image_path \\t min_ic_size\n",
     "\n",
     "imglist = 'imagelist.txt'\n",
     "\n",
     "with open(imglist, 'r') as f:\n",
-    "    lol = [x.strip().split(',') for x in f]\n",
+    "    lol = [x.strip().split('\\t') for x in f]\n",
     "\n",
     "print lol"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": null,
-   "metadata": {
-    "collapsed": true
-   },
-   "outputs": [],
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Computing bounding boxes\n"
+     ]
+    }
+   ],
    "source": [
     "# Initialize the detector\n",
     "icDetector = cd.componentDetector(\"ic\", \"gpu\")\n",
@@ -73,6 +83,15 @@
     "    helperModules.drawBoundingBoxes(im, boundingBoxes, \"Detected Image\")\n",
     "    "
    ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "collapsed": true
+   },
+   "outputs": [],
+   "source": []
   }
  ],
  "metadata": {
@@ -91,7 +110,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython2",
-   "version": "2.7.13"
+   "version": "2.7.6"
   }
  },
  "nbformat": 4,

setup.py

@@ -1,8 +1,8 @@
-#from distutils.core import setup
-from setuptools import setup, find_packages
+from distutils.core import setup
+#from setuptools import setup, find_packages
 import os
 
-datadir = 'models'
+datadir = 'detector/models'
 datafiles = [(d, [os.path.join(d,f) for f in files])
     for d, folders, files in os.walk(datadir)]
 
@@ -15,9 +15,10 @@
     version='0.1dev0',
     author='Gayathri Mahalingam, Kevin Marshall Gay',
     author_email='[email protected]',
-    packages=find_packages(),
-    package_data={'detector':datafiles},
+    packages=['detector'],
+    package_data={'detector':['models/ic/*', 'models/resistor/*', 'models/capacitor/*']},
     include_package_data=True,
+    data_files=datafiles,
     install_requires=required,
     setup_requires=[],
     tests_require=[],