# 机器学习中，使用NMS对框取优

# http://www.pyimagesearch.com/2015/02/16/faster-non-maximum-suppression-python/

import numpy as np

class NMSuppression(object):
def __init__(self, bbs, overlapThreshold = 0.45):
self.bbs = bbs
self.overlapThreshold = overlapThreshold

def _check_empty(self):
# return an empty list, if there are no boxes
if len(self.bbs) == 0:
return []
else:
return self.bbs

def _check_dtype(self):
# if the bounding boxes integers, convert them to floats (divisions)
if self.bbs.dtype.kind == "i":
self.bbs = self.bbs.astype("float")
return self.bbs

def bb_coordinates(self):
# get the coordinates of the bounding boxes
x1 = self.bbs[:, 0]
y1 = self.bbs[:, 1]
x2 = self.bbs[:, 2]
y2 = self.bbs[:, 3]
return x1, y1, x2, y2

def bb_area(self):
# compute the area of the bounding boxes
x1, y1, x2, y2 = self.bb_coordinates()
area = (x2 - x1 + 1) * (y2 - y1 + 1)
return area

def calc_ovarlap(self, x1, y1, x2, y2, idxs, last, i, area):
# 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]])
xx2 = np.minimum(x2[i], x2[idxs[:last]])
yy2 = np.minimum(y2[i], y2[idxs[:last]])

# compute the width and height of the bounding box
w = np.maximum(0, xx2 - xx1 + 1)
h = np.maximum(0, yy2 - yy1 + 1)

# compute the ratio of overlap
overlap = (w * h) / area[idxs[:last]]

return overlap

def slow_suppress(self):
self._check_empty()
self._check_dtype()

# initialize the list of picked indexes
picked = []

x1, y1, x2, y2 = self.bb_coordinates()

# compute the area of the bounding boxes
area = self.bb_area()

# sort the bounding boxes by the bottom-right y-coordinate of the bounding box
idxs = np.argsort(y2)

# keep looping while some indexes still remain in the indexes list
while len(idxs) > 0:
# grab the last index in the indexes list, add the index
# value to the list of picked indexes, then initialize
# the suppression list (i.e. indexes that will be deleted)
# using the last index
last = len(idxs) - 1
i = idxs[last]
picked.append(i)
suppress = [last]

# loop over all indexes in the indexes list
for pos in xrange(0, last):
# grab the current index
j = idxs[pos]

# 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 = max(x1[i], x1[j])
yy1 = max(y1[i], y1[j])
xx2 = min(x2[i], x2[j])
yy2 = min(y2[i], y2[j])

# compute the width and height of the bounding box
w = max(0, xx2 - xx1 + 1)
h = max(0, yy2 - yy1 + 1)

# compute the ratio of overlap between the computed
# bounding box and the bounding box in the area list
overlap = float(w * h) / area[j]

# if there is sufficient overlap, suppress the
# current bounding box
if overlap > self.overlapThreshold:
suppress.append(pos)

# delete all indexes from the index list that are in the
# suppression list
idxs = np.delete(idxs, suppress)

# return only the bounding boxes that were picked
return self.bbs[picked]

def fast_suppress(self):
self._check_empty()
self._check_dtype()

# initialize the list of picked indexes
picked = []

x1, y1, x2, y2 = self.bb_coordinates()

# compute the area of the bounding boxes
area = self.bb_area()

# sort the bounding boxes by the bottom-right y-coordinate of the bounding box
idxs = np.argsort(y2)

# keep looping while some indexes still remain in the indexes list
while len(idxs) > 0:
# take the last index in the indexes list and add the
# index value to the list of picked indexes
last = len(idxs) - 1
i = idxs[last]
picked.append(i)

overlap = self.calc_ovarlap(x1, y1, x2, y2, idxs, last, i, area)

# delete all indexes from the index list that have
idxs = np.delete(idxs, np.concatenate(([last], np.where(overlap > self.overlapThreshold)[0])))

# return only the bounding boxes that were picked using the
# integer data type

return self.bbs[picked].astype("int")

#taken from: http://www.pyimagesearch.com/2014/11/17/non-maximum-suppression-object-detection-python
"""
Project parts (taken from the tutorial above):
1.    Sampling positive images
2.    Sampling negative images
3.    Training a Linear SVM
4.    Performing hard-negative mining
5.    Re-training your Linear SVM using the hard-negative samples
suppression to ignore redundant, overlapping bounding boxes

The sample images in this project are taken from the web (labeled as: no licensing needed for non-comertial use).
"""

from nm_suppression import NMSuppression
import numpy as np
import cv2

# construct a list containing the images that will be examined
# along with their respective bounding boxes
images = [
("images/africa.jpeg", np.array([
(12, 84, 140, 212),
(24, 84, 152, 212),
(36, 84, 164, 212),
(12, 96, 140, 224),
(24, 96, 152, 224),
(24, 108, 152, 236)])),
("images/girl.jpeg", np.array([
(114, 60, 178, 124),
(120, 60, 184, 124),
(114, 66, 178, 130)])),
("images/monroe.jpeg", np.array([
(12, 30, 76, 94),
(12, 36, 76, 100),
(72, 36, 200, 164),
(84, 48, 212, 176)]))]

# loop over the images
for (imagePath, boundingBoxes) in images:
# load the image and clone it
print "%d initial bounding boxes" % (len(boundingBoxes))
orig = image.copy()

# loop over the bounding boxes for each image and draw them
for (startX, startY, endX, endY) in boundingBoxes:
cv2.rectangle(orig, (startX, startY), (endX, endY), (0, 0, 255), 2)

# perform non-maximum suppression on the bounding boxes
p = NMSuppression(bbs=boundingBoxes, overlapThreshold=0.5)
pick = p.fast_suppress()
print "after applying non-maximum, %d bounding boxes" % (len(pick))

# loop over the picked bounding boxes and draw them
for (startX, startY, endX, endY) in pick:
cv2.rectangle(image, (startX, startY), (endX, endY), (0, 255, 0), 2)

# display the images
cv2.imshow("Original", orig)
cv2.imshow("After NMS", image)
cv2.waitKey(0)

0 条回应