11from BoundingBox import *
22from utils import *
33
4- class BoundingBoxes :
54
5+ class BoundingBoxes :
66 def __init__ (self ):
77 self ._boundingBoxes = []
88
@@ -11,20 +11,20 @@ def addBoundingBox(self, bb):
1111
1212 def removeBoundingBox (self , _boundingBox ):
1313 for d in self ._boundingBoxes :
14- if BoundingBox .compare (d ,_boundingBox ):
14+ if BoundingBox .compare (d , _boundingBox ):
1515 del self ._boundingBoxes [d ]
1616 return
17-
17+
1818 def removeAllBoundingBoxes (self ):
1919 self ._boundingBoxes = []
20-
20+
2121 def getBoundingBoxes (self ):
2222 return self ._boundingBoxes
2323
2424 def getBoundingBoxByClass (self , classId ):
2525 boundingBoxes = []
2626 for d in self ._boundingBoxes :
27- if d .getClassId () == classId : # get only specified bounding box type
27+ if d .getClassId () == classId : # get only specified bounding box type
2828 boundingBoxes .append (d )
2929 return boundingBoxes
3030
@@ -38,21 +38,21 @@ def getClasses(self):
3838
3939 def getBoundingBoxesByType (self , bbType ):
4040 # get only specified bb type
41- return [d for d in self ._boundingBoxes if d .getBBType () == bbType ]
41+ return [d for d in self ._boundingBoxes if d .getBBType () == bbType ]
4242
4343 def getBoundingBoxesByImageName (self , imageName ):
4444 # get only specified bb type
45- return [d for d in self ._boundingBoxes if d .getImageName () == imageName ]
45+ return [d for d in self ._boundingBoxes if d .getImageName () == imageName ]
4646
4747 def count (self , bbType = None ):
48- if bbType == None : # Return all bounding boxes
48+ if bbType is None : # Return all bounding boxes
4949 return len (self ._boundingBoxes )
5050 count = 0
5151 for d in self ._boundingBoxes :
52- if d .getBBType () == bbType : # get only specified bb type
52+ if d .getBBType () == bbType : # get only specified bb type
5353 count += 1
5454 return count
55-
55+
5656 def clone (self ):
5757 newBoundingBoxes = BoundingBoxes ()
5858 for d in self ._boundingBoxes :
@@ -63,116 +63,15 @@ def clone(self):
6363 def drawAllBoundingBoxes (self , image , imageName ):
6464 bbxes = self .getBoundingBoxesByImageName (imageName )
6565 for bb in bbxes :
66- if bb .getBBType () == BBType .GroundTruth : # if ground truth
67- image = add_bb_into_image (image , bb , color = (0 ,255 ,0 )) # green
68- else : # if detection
69- image = add_bb_into_image (image , bb , color = (255 ,0 , 0 )) # red
70- return image
66+ if bb .getBBType () == BBType .GroundTruth : # if ground truth
67+ image = add_bb_into_image (image , bb , color = (0 , 255 , 0 )) # green
68+ else : # if detection
69+ image = add_bb_into_image (image , bb , color = (255 , 0 , 0 )) # red
70+ return image
7171
7272 # def drawAllBoundingBoxes(self, image):
7373 # for gt in self.getBoundingBoxesByType(BBType.GroundTruth):
7474 # image = add_bb_into_image(image, gt ,color=(0,255,0))
7575 # for det in self.getBoundingBoxesByType(BBType.Detected):
7676 # image = add_bb_into_image(image, det ,color=(255,0,0))
77- # return image
78-
79- # @staticmethod
80- # def evaluateDetections(gtDetection, evalDetection, minIoUTruePos=0.0):
81- # #####################################################################
82- # # Evaluate Intersection over Union (IoU):
83- # #
84- # # - Only detected objects that are overlapped with the same class
85- # # groundtruth objects will be taken into account (e.g. a "cat"
86- # # can only be compared to another "cat").
87- # # - Among multiple detections for a unique groundtruth bounding box,
88- # # only the one with the highest IoU will be taken into consid-
89- # # eration.
90- # # - As the highest IoU is taken into consideration, the bounding box
91- # # pair that is considered a match (one from groundtruth and the
92- # # other is the detected one), is removed. This way these bound-
93- # # ing boxes won't be considered in further checking for this
94- # # image. Then only maximum IoU will be added to IoU_sum for
95- # # the image.
96- # # - The IoU of an image pair is the average of its IoU.
97- # # - IoU of the image = IoU_sum / (True_positives + False_positives) which
98- # # is the same as average(IoU_sum)
99- # #
100- # #####################################################################
101- # # Evaluate True Positive (TP) and False Positive (FP):
102- # #
103- # # - If detected "cat" isn't overlaped with "cat" (or overlaped with "dog"),
104- # # then this is false_positive and its IoU = 0
105- # # - All non-maximum IoUs = 0 and they are false_positives.
106- # # - If a detected "cat" is somehow overlapped with a "cat", then it is
107- # # accounted as a True Positive and no other detection will be
108- # # considered for those bounding boxes.
109- # #####################################################################
110- # IoUs = []
111- # # Initiate True Positives and False Positive counts
112- # TP = 0
113- # FP = 0
114- # FN = 0
115- # # Detections.Teste(gtDetection, evalDetection)
116- # dets = Detections.SeparateClasses(gtDetection, evalDetection)
117-
118-
119- # # Multiple detections of the same object in an image are considered
120- # # false detections e.g. 5 detections of a single object is counted as 1 correct detec-
121- # # tion and 4 false detections – it is the responsibility of the participant’s system
122- # # to filter multiple detections from its output
123-
124- # # For each class, get its GTs and detections
125- # for d in dets:
126- # gtDetections = d[1]
127- # evalDetections = d[2]
128-
129- # # for each evalDetection, find the best (lowest IOU) gtDetection
130- # # note: the eval detection must be the same class
131- # while len(gtDetections) > 0:
132- # for detEval in evalDetections:
133- # bb = detEval.getAbsoluteBoundingBox()
134- # bestIoU = 0
135- # bestGT = None
136- # # find the bb with lowest IOU
137- # for detGT in gtDetections:
138- # iou = YOLOHelper.iou(bb, detGT.getAbsoluteBoundingBox())
139-
140- # # Show blank image with the bounding boxes
141- # img = np.zeros((detGT.height_img,detGT.width_img,3), np.uint8)
142- # aa = detEval.getAbsoluteBoundingBox()
143- # img = cv2.rectangle(img, (aa[0],aa[1]), (aa[2],aa[3]), (0,0,255), 6)
144- # for de in evalDetection.detections:
145- # aaa = de.getAbsoluteBoundingBox()
146- # img = cv2.rectangle(img, (aaa[0],aaa[1]), (aaa[2],aaa[3]), (0,0,255), 2)
147- # bbb = detGT.getAbsoluteBoundingBox()
148- # img = cv2.rectangle(img, (bbb[0],bbb[1]), (bbb[2],bbb[3]), (0,255,0), 6)
149- # for gt in gtDetection.detections:
150- # bbb = gt.getAbsoluteBoundingBox()
151- # img = cv2.rectangle(img, (bbb[0],bbb[1]), (bbb[2],bbb[3]), (0,255,0), 2)
152- # cv2.imshow("IoU %.2f" % iou,img)
153- # cv2.waitKey(0)
154- # cv2.destroyWindow("IoU %.2f" % iou)
155-
156- # if iou > 1 or iou < 0:
157- # raise ValueError('IOU value out of limits: %f' % iou)
158-
159- # if iou > bestIoU:
160- # bestGT = detGT
161- # bestIoU = iou
162-
163- # IoUs.append(bestIoU)
164- # # Increment False Positives or True Positives
165- # if bestIoU == 0: # Detection makes no overlap with any object: it is a false positive
166- # FP = FP + 1
167- # elif (bestIoU >= minIoUTruePos):
168- # TP = TP + 1
169- # else: #bestIoU < minIoUTruePos : The overlapped IOU is below the threshold
170- # FP = FP + 1 # Detection is a false positive
171-
172- # # Now remove this detected BB and go to the next
173- # evalDetection.detections.remove(detEval)
174- # continue
175-
176- # FN = FN + len(gtDetection.detections)
177- # # Return average IoU among all detected bounding boxes, True Positives and False Positives
178- # return FN, TP, FP
77+ # return image
0 commit comments