mirror of
https://github.com/opencv/opencv.git
synced 2026-07-29 23:33:05 +04:00
Merge pull request #25017 from kaingwade:ml_to_contrib
Move ml to opencv_contrib #25017 OpenCV cleanup: #24997 opencv_contrib: opencv/opencv_contrib#3636
This commit is contained in:
WU Jia
GitHub
@@ -1,194 +0,0 @@
|
||||
#!/usr/bin/env python
|
||||
|
||||
'''
|
||||
SVM and KNearest digit recognition.
|
||||
|
||||
Sample loads a dataset of handwritten digits from 'digits.png'.
|
||||
Then it trains a SVM and KNearest classifiers on it and evaluates
|
||||
their accuracy.
|
||||
|
||||
Following preprocessing is applied to the dataset:
|
||||
- Moment-based image deskew (see deskew())
|
||||
- Digit images are split into 4 10x10 cells and 16-bin
|
||||
histogram of oriented gradients is computed for each
|
||||
cell
|
||||
- Transform histograms to space with Hellinger metric (see [1] (RootSIFT))
|
||||
|
||||
|
||||
[1] R. Arandjelovic, A. Zisserman
|
||||
"Three things everyone should know to improve object retrieval"
|
||||
http://www.robots.ox.ac.uk/~vgg/publications/2012/Arandjelovic12/arandjelovic12.pdf
|
||||
|
||||
Usage:
|
||||
digits.py
|
||||
'''
|
||||
|
||||
|
||||
# Python 2/3 compatibility
|
||||
from __future__ import print_function
|
||||
|
||||
import numpy as np
|
||||
import cv2 as cv
|
||||
|
||||
# built-in modules
|
||||
from multiprocessing.pool import ThreadPool
|
||||
|
||||
from numpy.linalg import norm
|
||||
|
||||
# local modules
|
||||
from common import clock, mosaic
|
||||
|
||||
|
||||
|
||||
SZ = 20 # size of each digit is SZ x SZ
|
||||
CLASS_N = 10
|
||||
DIGITS_FN = 'digits.png'
|
||||
|
||||
def split2d(img, cell_size, flatten=True):
|
||||
h, w = img.shape[:2]
|
||||
sx, sy = cell_size
|
||||
cells = [np.hsplit(row, w//sx) for row in np.vsplit(img, h//sy)]
|
||||
cells = np.array(cells)
|
||||
if flatten:
|
||||
cells = cells.reshape(-1, sy, sx)
|
||||
return cells
|
||||
|
||||
def load_digits(fn):
|
||||
fn = cv.samples.findFile(fn)
|
||||
print('loading "%s" ...' % fn)
|
||||
digits_img = cv.imread(fn, cv.IMREAD_GRAYSCALE)
|
||||
digits = split2d(digits_img, (SZ, SZ))
|
||||
labels = np.repeat(np.arange(CLASS_N), len(digits)/CLASS_N)
|
||||
return digits, labels
|
||||
|
||||
def deskew(img):
|
||||
m = cv.moments(img)
|
||||
if abs(m['mu02']) < 1e-2:
|
||||
return img.copy()
|
||||
skew = m['mu11']/m['mu02']
|
||||
M = np.float32([[1, skew, -0.5*SZ*skew], [0, 1, 0]])
|
||||
img = cv.warpAffine(img, M, (SZ, SZ), flags=cv.WARP_INVERSE_MAP | cv.INTER_LINEAR)
|
||||
return img
|
||||
|
||||
|
||||
class KNearest(object):
|
||||
def __init__(self, k = 3):
|
||||
self.k = k
|
||||
self.model = cv.ml.KNearest_create()
|
||||
|
||||
def train(self, samples, responses):
|
||||
self.model.train(samples, cv.ml.ROW_SAMPLE, responses)
|
||||
|
||||
def predict(self, samples):
|
||||
_retval, results, _neigh_resp, _dists = self.model.findNearest(samples, self.k)
|
||||
return results.ravel()
|
||||
|
||||
def load(self, fn):
|
||||
self.model = cv.ml.KNearest_load(fn)
|
||||
|
||||
def save(self, fn):
|
||||
self.model.save(fn)
|
||||
|
||||
class SVM(object):
|
||||
def __init__(self, C = 1, gamma = 0.5):
|
||||
self.model = cv.ml.SVM_create()
|
||||
self.model.setGamma(gamma)
|
||||
self.model.setC(C)
|
||||
self.model.setKernel(cv.ml.SVM_RBF)
|
||||
self.model.setType(cv.ml.SVM_C_SVC)
|
||||
|
||||
def train(self, samples, responses):
|
||||
self.model.train(samples, cv.ml.ROW_SAMPLE, responses)
|
||||
|
||||
def predict(self, samples):
|
||||
return self.model.predict(samples)[1].ravel()
|
||||
|
||||
def load(self, fn):
|
||||
self.model = cv.ml.SVM_load(fn)
|
||||
|
||||
def save(self, fn):
|
||||
self.model.save(fn)
|
||||
|
||||
def evaluate_model(model, digits, samples, labels):
|
||||
resp = model.predict(samples)
|
||||
err = (labels != resp).mean()
|
||||
print('error: %.2f %%' % (err*100))
|
||||
|
||||
confusion = np.zeros((10, 10), np.int32)
|
||||
for i, j in zip(labels, resp):
|
||||
confusion[i, int(j)] += 1
|
||||
print('confusion matrix:')
|
||||
print(confusion)
|
||||
print()
|
||||
|
||||
vis = []
|
||||
for img, flag in zip(digits, resp == labels):
|
||||
img = cv.cvtColor(img, cv.COLOR_GRAY2BGR)
|
||||
if not flag:
|
||||
img[...,:2] = 0
|
||||
vis.append(img)
|
||||
return mosaic(25, vis)
|
||||
|
||||
def preprocess_simple(digits):
|
||||
return np.float32(digits).reshape(-1, SZ*SZ) / 255.0
|
||||
|
||||
def preprocess_hog(digits):
|
||||
samples = []
|
||||
for img in digits:
|
||||
gx = cv.Sobel(img, cv.CV_32F, 1, 0)
|
||||
gy = cv.Sobel(img, cv.CV_32F, 0, 1)
|
||||
mag, ang = cv.cartToPolar(gx, gy)
|
||||
bin_n = 16
|
||||
bin = np.int32(bin_n*ang/(2*np.pi))
|
||||
bin_cells = bin[:10,:10], bin[10:,:10], bin[:10,10:], bin[10:,10:]
|
||||
mag_cells = mag[:10,:10], mag[10:,:10], mag[:10,10:], mag[10:,10:]
|
||||
hists = [np.bincount(b.ravel(), m.ravel(), bin_n) for b, m in zip(bin_cells, mag_cells)]
|
||||
hist = np.hstack(hists)
|
||||
|
||||
# transform to Hellinger kernel
|
||||
eps = 1e-7
|
||||
hist /= hist.sum() + eps
|
||||
hist = np.sqrt(hist)
|
||||
hist /= norm(hist) + eps
|
||||
|
||||
samples.append(hist)
|
||||
return np.float32(samples)
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
print(__doc__)
|
||||
|
||||
digits, labels = load_digits(DIGITS_FN)
|
||||
|
||||
print('preprocessing...')
|
||||
# shuffle digits
|
||||
rand = np.random.RandomState(321)
|
||||
shuffle = rand.permutation(len(digits))
|
||||
digits, labels = digits[shuffle], labels[shuffle]
|
||||
|
||||
digits2 = list(map(deskew, digits))
|
||||
samples = preprocess_hog(digits2)
|
||||
|
||||
train_n = int(0.9*len(samples))
|
||||
cv.imshow('test set', mosaic(25, digits[train_n:]))
|
||||
digits_train, digits_test = np.split(digits2, [train_n])
|
||||
samples_train, samples_test = np.split(samples, [train_n])
|
||||
labels_train, labels_test = np.split(labels, [train_n])
|
||||
|
||||
|
||||
print('training KNearest...')
|
||||
model = KNearest(k=4)
|
||||
model.train(samples_train, labels_train)
|
||||
vis = evaluate_model(model, digits_test, samples_test, labels_test)
|
||||
cv.imshow('KNearest test', vis)
|
||||
|
||||
print('training SVM...')
|
||||
model = SVM(C=2.67, gamma=5.383)
|
||||
model.train(samples_train, labels_train)
|
||||
vis = evaluate_model(model, digits_test, samples_test, labels_test)
|
||||
cv.imshow('SVM test', vis)
|
||||
print('saving SVM as "digits_svm.dat"...')
|
||||
model.save('digits_svm.dat')
|
||||
|
||||
cv.waitKey(0)
|
||||
cv.destroyAllWindows()
|
||||
@@ -1,132 +0,0 @@
|
||||
#!/usr/bin/env python
|
||||
|
||||
'''
|
||||
Digit recognition adjustment.
|
||||
Grid search is used to find the best parameters for SVM and KNearest classifiers.
|
||||
SVM adjustment follows the guidelines given in
|
||||
http://www.csie.ntu.edu.tw/~cjlin/papers/guide/guide.pdf
|
||||
|
||||
Usage:
|
||||
digits_adjust.py [--model {svm|knearest}]
|
||||
|
||||
--model {svm|knearest} - select the classifier (SVM is the default)
|
||||
|
||||
'''
|
||||
|
||||
import numpy as np
|
||||
import cv2 as cv
|
||||
|
||||
from multiprocessing.pool import ThreadPool
|
||||
|
||||
from digits import *
|
||||
|
||||
def cross_validate(model_class, params, samples, labels, kfold = 3, pool = None):
|
||||
n = len(samples)
|
||||
folds = np.array_split(np.arange(n), kfold)
|
||||
def f(i):
|
||||
model = model_class(**params)
|
||||
test_idx = folds[i]
|
||||
train_idx = list(folds)
|
||||
train_idx.pop(i)
|
||||
train_idx = np.hstack(train_idx)
|
||||
train_samples, train_labels = samples[train_idx], labels[train_idx]
|
||||
test_samples, test_labels = samples[test_idx], labels[test_idx]
|
||||
model.train(train_samples, train_labels)
|
||||
resp = model.predict(test_samples)
|
||||
score = (resp != test_labels).mean()
|
||||
print(".", end='')
|
||||
return score
|
||||
if pool is None:
|
||||
scores = list(map(f, range(kfold)))
|
||||
else:
|
||||
scores = pool.map(f, range(kfold))
|
||||
return np.mean(scores)
|
||||
|
||||
|
||||
class App(object):
|
||||
def __init__(self):
|
||||
self._samples, self._labels = self.preprocess()
|
||||
|
||||
def preprocess(self):
|
||||
digits, labels = load_digits(DIGITS_FN)
|
||||
shuffle = np.random.permutation(len(digits))
|
||||
digits, labels = digits[shuffle], labels[shuffle]
|
||||
digits2 = list(map(deskew, digits))
|
||||
samples = preprocess_hog(digits2)
|
||||
return samples, labels
|
||||
|
||||
def get_dataset(self):
|
||||
return self._samples, self._labels
|
||||
|
||||
def run_jobs(self, f, jobs):
|
||||
pool = ThreadPool(processes=cv.getNumberOfCPUs())
|
||||
ires = pool.imap_unordered(f, jobs)
|
||||
return ires
|
||||
|
||||
def adjust_SVM(self):
|
||||
Cs = np.logspace(0, 10, 15, base=2)
|
||||
gammas = np.logspace(-7, 4, 15, base=2)
|
||||
scores = np.zeros((len(Cs), len(gammas)))
|
||||
scores[:] = np.nan
|
||||
|
||||
print('adjusting SVM (may take a long time) ...')
|
||||
def f(job):
|
||||
i, j = job
|
||||
samples, labels = self.get_dataset()
|
||||
params = dict(C = Cs[i], gamma=gammas[j])
|
||||
score = cross_validate(SVM, params, samples, labels)
|
||||
return i, j, score
|
||||
|
||||
ires = self.run_jobs(f, np.ndindex(*scores.shape))
|
||||
for count, (i, j, score) in enumerate(ires):
|
||||
scores[i, j] = score
|
||||
print('%d / %d (best error: %.2f %%, last: %.2f %%)' %
|
||||
(count+1, scores.size, np.nanmin(scores)*100, score*100))
|
||||
print(scores)
|
||||
|
||||
print('writing score table to "svm_scores.npz"')
|
||||
np.savez('svm_scores.npz', scores=scores, Cs=Cs, gammas=gammas)
|
||||
|
||||
i, j = np.unravel_index(scores.argmin(), scores.shape)
|
||||
best_params = dict(C = Cs[i], gamma=gammas[j])
|
||||
print('best params:', best_params)
|
||||
print('best error: %.2f %%' % (scores.min()*100))
|
||||
return best_params
|
||||
|
||||
def adjust_KNearest(self):
|
||||
print('adjusting KNearest ...')
|
||||
def f(k):
|
||||
samples, labels = self.get_dataset()
|
||||
err = cross_validate(KNearest, dict(k=k), samples, labels)
|
||||
return k, err
|
||||
best_err, best_k = np.inf, -1
|
||||
for k, err in self.run_jobs(f, range(1, 9)):
|
||||
if err < best_err:
|
||||
best_err, best_k = err, k
|
||||
print('k = %d, error: %.2f %%' % (k, err*100))
|
||||
best_params = dict(k=best_k)
|
||||
print('best params:', best_params, 'err: %.2f' % (best_err*100))
|
||||
return best_params
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
import getopt
|
||||
import sys
|
||||
|
||||
print(__doc__)
|
||||
|
||||
args, _ = getopt.getopt(sys.argv[1:], '', ['model='])
|
||||
args = dict(args)
|
||||
args.setdefault('--model', 'svm')
|
||||
args.setdefault('--env', '')
|
||||
if args['--model'] not in ['svm', 'knearest']:
|
||||
print('unknown model "%s"' % args['--model'])
|
||||
sys.exit(1)
|
||||
|
||||
t = clock()
|
||||
app = App()
|
||||
if args['--model'] == 'knearest':
|
||||
app.adjust_KNearest()
|
||||
else:
|
||||
app.adjust_SVM()
|
||||
print('work time: %f s' % (clock() - t))
|
||||
@@ -1,109 +0,0 @@
|
||||
#!/usr/bin/env python
|
||||
'''
|
||||
Digit recognition from video.
|
||||
|
||||
Run digits.py before, to train and save the SVM.
|
||||
|
||||
Usage:
|
||||
digits_video.py [{camera_id|video_file}]
|
||||
'''
|
||||
|
||||
# Python 2/3 compatibility
|
||||
from __future__ import print_function
|
||||
|
||||
import numpy as np
|
||||
import cv2 as cv
|
||||
|
||||
# built-in modules
|
||||
import os
|
||||
import sys
|
||||
|
||||
# local modules
|
||||
import video
|
||||
from common import mosaic
|
||||
|
||||
from digits import *
|
||||
|
||||
def main():
|
||||
try:
|
||||
src = sys.argv[1]
|
||||
except:
|
||||
src = 0
|
||||
cap = video.create_capture(src, fallback='synth:bg={}:noise=0.05'.format(cv.samples.findFile('sudoku.png')))
|
||||
|
||||
classifier_fn = 'digits_svm.dat'
|
||||
if not os.path.exists(classifier_fn):
|
||||
print('"%s" not found, run digits.py first' % classifier_fn)
|
||||
return
|
||||
|
||||
model = cv.ml.SVM_load(classifier_fn)
|
||||
|
||||
while True:
|
||||
_ret, frame = cap.read()
|
||||
gray = cv.cvtColor(frame, cv.COLOR_BGR2GRAY)
|
||||
|
||||
|
||||
bin = cv.adaptiveThreshold(gray, 255, cv.ADAPTIVE_THRESH_MEAN_C, cv.THRESH_BINARY_INV, 31, 10)
|
||||
bin = cv.medianBlur(bin, 3)
|
||||
contours, heirs = cv.findContours( bin.copy(), cv.RETR_CCOMP, cv.CHAIN_APPROX_SIMPLE)
|
||||
try:
|
||||
heirs = heirs[0]
|
||||
except:
|
||||
heirs = []
|
||||
|
||||
for cnt, heir in zip(contours, heirs):
|
||||
_, _, _, outer_i = heir
|
||||
if outer_i >= 0:
|
||||
continue
|
||||
x, y, w, h = cv.boundingRect(cnt)
|
||||
if not (16 <= h <= 64 and w <= 1.2*h):
|
||||
continue
|
||||
pad = max(h-w, 0)
|
||||
x, w = x - (pad // 2), w + pad
|
||||
cv.rectangle(frame, (x, y), (x+w, y+h), (0, 255, 0))
|
||||
|
||||
bin_roi = bin[y:,x:][:h,:w]
|
||||
|
||||
m = bin_roi != 0
|
||||
if not 0.1 < m.mean() < 0.4:
|
||||
continue
|
||||
'''
|
||||
gray_roi = gray[y:,x:][:h,:w]
|
||||
v_in, v_out = gray_roi[m], gray_roi[~m]
|
||||
if v_out.std() > 10.0:
|
||||
continue
|
||||
s = "%f, %f" % (abs(v_in.mean() - v_out.mean()), v_out.std())
|
||||
cv.putText(frame, s, (x, y), cv.FONT_HERSHEY_PLAIN, 1.0, (200, 0, 0), thickness = 1)
|
||||
'''
|
||||
|
||||
s = 1.5*float(h)/SZ
|
||||
m = cv.moments(bin_roi)
|
||||
c1 = np.float32([m['m10'], m['m01']]) / m['m00']
|
||||
c0 = np.float32([SZ/2, SZ/2])
|
||||
t = c1 - s*c0
|
||||
A = np.zeros((2, 3), np.float32)
|
||||
A[:,:2] = np.eye(2)*s
|
||||
A[:,2] = t
|
||||
bin_norm = cv.warpAffine(bin_roi, A, (SZ, SZ), flags=cv.WARP_INVERSE_MAP | cv.INTER_LINEAR)
|
||||
bin_norm = deskew(bin_norm)
|
||||
if x+w+SZ < frame.shape[1] and y+SZ < frame.shape[0]:
|
||||
frame[y:,x+w:][:SZ, :SZ] = bin_norm[...,np.newaxis]
|
||||
|
||||
sample = preprocess_hog([bin_norm])
|
||||
digit = model.predict(sample)[1].ravel()
|
||||
cv.putText(frame, '%d'%digit, (x, y), cv.FONT_HERSHEY_PLAIN, 1.0, (200, 0, 0), thickness = 1)
|
||||
|
||||
|
||||
cv.imshow('frame', frame)
|
||||
cv.imshow('bin', bin)
|
||||
ch = cv.waitKey(1)
|
||||
if ch == 27:
|
||||
break
|
||||
|
||||
print('Done')
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
print(__doc__)
|
||||
main()
|
||||
cv.destroyAllWindows()
|
||||
@@ -1,69 +0,0 @@
|
||||
#!/usr/bin/env python
|
||||
|
||||
import numpy as np
|
||||
import cv2 as cv
|
||||
|
||||
from numpy import random
|
||||
|
||||
def make_gaussians(cluster_n, img_size):
|
||||
points = []
|
||||
ref_distrs = []
|
||||
for _i in range(cluster_n):
|
||||
mean = (0.1 + 0.8*random.rand(2)) * img_size
|
||||
a = (random.rand(2, 2)-0.5)*img_size*0.1
|
||||
cov = np.dot(a.T, a) + img_size*0.05*np.eye(2)
|
||||
n = 100 + random.randint(900)
|
||||
pts = random.multivariate_normal(mean, cov, n)
|
||||
points.append( pts )
|
||||
ref_distrs.append( (mean, cov) )
|
||||
points = np.float32( np.vstack(points) )
|
||||
return points, ref_distrs
|
||||
|
||||
def draw_gaussain(img, mean, cov, color):
|
||||
x, y = mean
|
||||
w, u, _vt = cv.SVDecomp(cov)
|
||||
ang = np.arctan2(u[1, 0], u[0, 0])*(180/np.pi)
|
||||
s1, s2 = np.sqrt(w)*3.0
|
||||
cv.ellipse(img, (int(x), int(y)), (int(s1), int(s2)), ang, 0, 360, color, 1, cv.LINE_AA)
|
||||
|
||||
|
||||
def main():
|
||||
cluster_n = 5
|
||||
img_size = 512
|
||||
|
||||
print('press any key to update distributions, ESC - exit\n')
|
||||
|
||||
while True:
|
||||
print('sampling distributions...')
|
||||
points, ref_distrs = make_gaussians(cluster_n, img_size)
|
||||
|
||||
print('EM (opencv) ...')
|
||||
em = cv.ml.EM_create()
|
||||
em.setClustersNumber(cluster_n)
|
||||
em.setCovarianceMatrixType(cv.ml.EM_COV_MAT_GENERIC)
|
||||
em.trainEM(points)
|
||||
means = em.getMeans()
|
||||
covs = em.getCovs() # Known bug: https://github.com/opencv/opencv/pull/4232
|
||||
found_distrs = zip(means, covs)
|
||||
print('ready!\n')
|
||||
|
||||
img = np.zeros((img_size, img_size, 3), np.uint8)
|
||||
for x, y in np.int32(points):
|
||||
cv.circle(img, (x, y), 1, (255, 255, 255), -1)
|
||||
for m, cov in ref_distrs:
|
||||
draw_gaussain(img, m, cov, (0, 255, 0))
|
||||
for m, cov in found_distrs:
|
||||
draw_gaussain(img, m, cov, (0, 0, 255))
|
||||
|
||||
cv.imshow('gaussian mixture', img)
|
||||
ch = cv.waitKey(0)
|
||||
if ch == 27:
|
||||
break
|
||||
|
||||
print('Done')
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
print(__doc__)
|
||||
main()
|
||||
cv.destroyAllWindows()
|
||||
@@ -1,194 +0,0 @@
|
||||
#!/usr/bin/env python
|
||||
|
||||
'''
|
||||
The sample demonstrates how to train Random Trees classifier
|
||||
(or Boosting classifier, or MLP, or Knearest, or Support Vector Machines) using the provided dataset.
|
||||
|
||||
We use the sample database letter-recognition.data
|
||||
from UCI Repository, here is the link:
|
||||
|
||||
Newman, D.J. & Hettich, S. & Blake, C.L. & Merz, C.J. (1998).
|
||||
UCI Repository of machine learning databases
|
||||
[http://www.ics.uci.edu/~mlearn/MLRepository.html].
|
||||
Irvine, CA: University of California, Department of Information and Computer Science.
|
||||
|
||||
The dataset consists of 20000 feature vectors along with the
|
||||
responses - capital latin letters A..Z.
|
||||
The first 10000 samples are used for training
|
||||
and the remaining 10000 - to test the classifier.
|
||||
======================================================
|
||||
USAGE:
|
||||
letter_recog.py [--model <model>]
|
||||
[--data <data fn>]
|
||||
[--load <model fn>] [--save <model fn>]
|
||||
|
||||
Models: RTrees, KNearest, Boost, SVM, MLP
|
||||
'''
|
||||
|
||||
# Python 2/3 compatibility
|
||||
from __future__ import print_function
|
||||
|
||||
import numpy as np
|
||||
import cv2 as cv
|
||||
|
||||
def load_base(fn):
|
||||
a = np.loadtxt(fn, np.float32, delimiter=',', converters={ 0 : lambda ch : ord(ch)-ord('A') })
|
||||
samples, responses = a[:,1:], a[:,0]
|
||||
return samples, responses
|
||||
|
||||
class LetterStatModel(object):
|
||||
class_n = 26
|
||||
train_ratio = 0.5
|
||||
|
||||
def load(self, fn):
|
||||
self.model = self.model.load(fn)
|
||||
def save(self, fn):
|
||||
self.model.save(fn)
|
||||
|
||||
def unroll_samples(self, samples):
|
||||
sample_n, var_n = samples.shape
|
||||
new_samples = np.zeros((sample_n * self.class_n, var_n+1), np.float32)
|
||||
new_samples[:,:-1] = np.repeat(samples, self.class_n, axis=0)
|
||||
new_samples[:,-1] = np.tile(np.arange(self.class_n), sample_n)
|
||||
return new_samples
|
||||
|
||||
def unroll_responses(self, responses):
|
||||
sample_n = len(responses)
|
||||
new_responses = np.zeros(sample_n*self.class_n, np.int32)
|
||||
resp_idx = np.int32( responses + np.arange(sample_n)*self.class_n )
|
||||
new_responses[resp_idx] = 1
|
||||
return new_responses
|
||||
|
||||
class RTrees(LetterStatModel):
|
||||
def __init__(self):
|
||||
self.model = cv.ml.RTrees_create()
|
||||
|
||||
def train(self, samples, responses):
|
||||
self.model.setMaxDepth(20)
|
||||
self.model.train(samples, cv.ml.ROW_SAMPLE, responses.astype(int))
|
||||
|
||||
def predict(self, samples):
|
||||
_ret, resp = self.model.predict(samples)
|
||||
return resp.ravel()
|
||||
|
||||
|
||||
class KNearest(LetterStatModel):
|
||||
def __init__(self):
|
||||
self.model = cv.ml.KNearest_create()
|
||||
|
||||
def train(self, samples, responses):
|
||||
self.model.train(samples, cv.ml.ROW_SAMPLE, responses)
|
||||
|
||||
def predict(self, samples):
|
||||
_retval, results, _neigh_resp, _dists = self.model.findNearest(samples, k = 10)
|
||||
return results.ravel()
|
||||
|
||||
|
||||
class Boost(LetterStatModel):
|
||||
def __init__(self):
|
||||
self.model = cv.ml.Boost_create()
|
||||
|
||||
def train(self, samples, responses):
|
||||
_sample_n, var_n = samples.shape
|
||||
new_samples = self.unroll_samples(samples)
|
||||
new_responses = self.unroll_responses(responses)
|
||||
var_types = np.array([cv.ml.VAR_NUMERICAL] * var_n + [cv.ml.VAR_CATEGORICAL, cv.ml.VAR_CATEGORICAL], np.uint8)
|
||||
|
||||
self.model.setWeakCount(15)
|
||||
self.model.setMaxDepth(10)
|
||||
self.model.train(cv.ml.TrainData_create(new_samples, cv.ml.ROW_SAMPLE, new_responses.astype(int), varType = var_types))
|
||||
|
||||
def predict(self, samples):
|
||||
new_samples = self.unroll_samples(samples)
|
||||
_ret, resp = self.model.predict(new_samples)
|
||||
|
||||
return resp.ravel().reshape(-1, self.class_n).argmax(1)
|
||||
|
||||
|
||||
class SVM(LetterStatModel):
|
||||
def __init__(self):
|
||||
self.model = cv.ml.SVM_create()
|
||||
|
||||
def train(self, samples, responses):
|
||||
self.model.setType(cv.ml.SVM_C_SVC)
|
||||
self.model.setC(1)
|
||||
self.model.setKernel(cv.ml.SVM_RBF)
|
||||
self.model.setGamma(.1)
|
||||
self.model.train(samples, cv.ml.ROW_SAMPLE, responses.astype(int))
|
||||
|
||||
def predict(self, samples):
|
||||
_ret, resp = self.model.predict(samples)
|
||||
return resp.ravel()
|
||||
|
||||
|
||||
class MLP(LetterStatModel):
|
||||
def __init__(self):
|
||||
self.model = cv.ml.ANN_MLP_create()
|
||||
|
||||
def train(self, samples, responses):
|
||||
_sample_n, var_n = samples.shape
|
||||
new_responses = self.unroll_responses(responses).reshape(-1, self.class_n)
|
||||
layer_sizes = np.int32([var_n, 100, 100, self.class_n])
|
||||
|
||||
self.model.setLayerSizes(layer_sizes)
|
||||
self.model.setTrainMethod(cv.ml.ANN_MLP_BACKPROP)
|
||||
self.model.setBackpropMomentumScale(0.0)
|
||||
self.model.setBackpropWeightScale(0.001)
|
||||
self.model.setTermCriteria((cv.TERM_CRITERIA_COUNT, 20, 0.01))
|
||||
self.model.setActivationFunction(cv.ml.ANN_MLP_SIGMOID_SYM, 2, 1)
|
||||
|
||||
self.model.train(samples, cv.ml.ROW_SAMPLE, np.float32(new_responses))
|
||||
|
||||
def predict(self, samples):
|
||||
_ret, resp = self.model.predict(samples)
|
||||
return resp.argmax(-1)
|
||||
|
||||
|
||||
|
||||
def main():
|
||||
import getopt
|
||||
import sys
|
||||
|
||||
models = [RTrees, KNearest, Boost, SVM, MLP] # NBayes
|
||||
models = dict( [(cls.__name__.lower(), cls) for cls in models] )
|
||||
|
||||
|
||||
args, dummy = getopt.getopt(sys.argv[1:], '', ['model=', 'data=', 'load=', 'save='])
|
||||
args = dict(args)
|
||||
args.setdefault('--model', 'svm')
|
||||
args.setdefault('--data', 'letter-recognition.data')
|
||||
|
||||
datafile = cv.samples.findFile(args['--data'])
|
||||
|
||||
print('loading data %s ...' % datafile)
|
||||
samples, responses = load_base(datafile)
|
||||
Model = models[args['--model']]
|
||||
model = Model()
|
||||
|
||||
train_n = int(len(samples)*model.train_ratio)
|
||||
if '--load' in args:
|
||||
fn = args['--load']
|
||||
print('loading model from %s ...' % fn)
|
||||
model.load(fn)
|
||||
else:
|
||||
print('training %s ...' % Model.__name__)
|
||||
model.train(samples[:train_n], responses[:train_n])
|
||||
|
||||
print('testing...')
|
||||
train_rate = np.mean(model.predict(samples[:train_n]) == responses[:train_n].astype(int))
|
||||
test_rate = np.mean(model.predict(samples[train_n:]) == responses[train_n:].astype(int))
|
||||
|
||||
print('train rate: %f test rate: %f' % (train_rate*100, test_rate*100))
|
||||
|
||||
if '--save' in args:
|
||||
fn = args['--save']
|
||||
print('saving model to %s ...' % fn)
|
||||
model.save(fn)
|
||||
|
||||
print('Done')
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
print(__doc__)
|
||||
main()
|
||||
cv.destroyAllWindows()
|
||||
@@ -1,62 +0,0 @@
|
||||
import cv2 as cv
|
||||
import numpy as np
|
||||
|
||||
# Set up training data
|
||||
## [setup1]
|
||||
labels = np.array([1, -1, -1, -1])
|
||||
trainingData = np.matrix([[501, 10], [255, 10], [501, 255], [10, 501]], dtype=np.float32)
|
||||
## [setup1]
|
||||
|
||||
# Train the SVM
|
||||
## [init]
|
||||
svm = cv.ml.SVM_create()
|
||||
svm.setType(cv.ml.SVM_C_SVC)
|
||||
svm.setKernel(cv.ml.SVM_LINEAR)
|
||||
svm.setTermCriteria((cv.TERM_CRITERIA_MAX_ITER, 100, 1e-6))
|
||||
## [init]
|
||||
## [train]
|
||||
svm.train(trainingData, cv.ml.ROW_SAMPLE, labels)
|
||||
## [train]
|
||||
|
||||
# Data for visual representation
|
||||
width = 512
|
||||
height = 512
|
||||
image = np.zeros((height, width, 3), dtype=np.uint8)
|
||||
|
||||
# Show the decision regions given by the SVM
|
||||
## [show]
|
||||
green = (0,255,0)
|
||||
blue = (255,0,0)
|
||||
for i in range(image.shape[0]):
|
||||
for j in range(image.shape[1]):
|
||||
sampleMat = np.matrix([[j,i]], dtype=np.float32)
|
||||
response = svm.predict(sampleMat)[1]
|
||||
|
||||
if response == 1:
|
||||
image[i,j] = green
|
||||
elif response == -1:
|
||||
image[i,j] = blue
|
||||
## [show]
|
||||
|
||||
# Show the training data
|
||||
## [show_data]
|
||||
thickness = -1
|
||||
cv.circle(image, (501, 10), 5, ( 0, 0, 0), thickness)
|
||||
cv.circle(image, (255, 10), 5, (255, 255, 255), thickness)
|
||||
cv.circle(image, (501, 255), 5, (255, 255, 255), thickness)
|
||||
cv.circle(image, ( 10, 501), 5, (255, 255, 255), thickness)
|
||||
## [show_data]
|
||||
|
||||
# Show support vectors
|
||||
## [show_vectors]
|
||||
thickness = 2
|
||||
sv = svm.getUncompressedSupportVectors()
|
||||
|
||||
for i in range(sv.shape[0]):
|
||||
cv.circle(image, (int(sv[i,0]), int(sv[i,1])), 6, (128, 128, 128), thickness)
|
||||
## [show_vectors]
|
||||
|
||||
cv.imwrite('result.png', image) # save the image
|
||||
|
||||
cv.imshow('SVM Simple Example', image) # show it to the user
|
||||
cv.waitKey()
|
||||
@@ -1,117 +0,0 @@
|
||||
from __future__ import print_function
|
||||
import cv2 as cv
|
||||
import numpy as np
|
||||
import random as rng
|
||||
|
||||
NTRAINING_SAMPLES = 100 # Number of training samples per class
|
||||
FRAC_LINEAR_SEP = 0.9 # Fraction of samples which compose the linear separable part
|
||||
|
||||
# Data for visual representation
|
||||
WIDTH = 512
|
||||
HEIGHT = 512
|
||||
I = np.zeros((HEIGHT, WIDTH, 3), dtype=np.uint8)
|
||||
|
||||
# --------------------- 1. Set up training data randomly ---------------------------------------
|
||||
trainData = np.empty((2*NTRAINING_SAMPLES, 2), dtype=np.float32)
|
||||
labels = np.empty((2*NTRAINING_SAMPLES, 1), dtype=np.int32)
|
||||
|
||||
rng.seed(100) # Random value generation class
|
||||
|
||||
# Set up the linearly separable part of the training data
|
||||
nLinearSamples = int(FRAC_LINEAR_SEP * NTRAINING_SAMPLES)
|
||||
|
||||
## [setup1]
|
||||
# Generate random points for the class 1
|
||||
trainClass = trainData[0:nLinearSamples,:]
|
||||
# The x coordinate of the points is in [0, 0.4)
|
||||
c = trainClass[:,0:1]
|
||||
c[:] = np.random.uniform(0.0, 0.4 * WIDTH, c.shape)
|
||||
# The y coordinate of the points is in [0, 1)
|
||||
c = trainClass[:,1:2]
|
||||
c[:] = np.random.uniform(0.0, HEIGHT, c.shape)
|
||||
|
||||
# Generate random points for the class 2
|
||||
trainClass = trainData[2*NTRAINING_SAMPLES-nLinearSamples:2*NTRAINING_SAMPLES,:]
|
||||
# The x coordinate of the points is in [0.6, 1]
|
||||
c = trainClass[:,0:1]
|
||||
c[:] = np.random.uniform(0.6*WIDTH, WIDTH, c.shape)
|
||||
# The y coordinate of the points is in [0, 1)
|
||||
c = trainClass[:,1:2]
|
||||
c[:] = np.random.uniform(0.0, HEIGHT, c.shape)
|
||||
## [setup1]
|
||||
|
||||
#------------------ Set up the non-linearly separable part of the training data ---------------
|
||||
## [setup2]
|
||||
# Generate random points for the classes 1 and 2
|
||||
trainClass = trainData[nLinearSamples:2*NTRAINING_SAMPLES-nLinearSamples,:]
|
||||
# The x coordinate of the points is in [0.4, 0.6)
|
||||
c = trainClass[:,0:1]
|
||||
c[:] = np.random.uniform(0.4*WIDTH, 0.6*WIDTH, c.shape)
|
||||
# The y coordinate of the points is in [0, 1)
|
||||
c = trainClass[:,1:2]
|
||||
c[:] = np.random.uniform(0.0, HEIGHT, c.shape)
|
||||
## [setup2]
|
||||
|
||||
#------------------------- Set up the labels for the classes ---------------------------------
|
||||
labels[0:NTRAINING_SAMPLES,:] = 1 # Class 1
|
||||
labels[NTRAINING_SAMPLES:2*NTRAINING_SAMPLES,:] = 2 # Class 2
|
||||
|
||||
#------------------------ 2. Set up the support vector machines parameters --------------------
|
||||
print('Starting training process')
|
||||
## [init]
|
||||
svm = cv.ml.SVM_create()
|
||||
svm.setType(cv.ml.SVM_C_SVC)
|
||||
svm.setC(0.1)
|
||||
svm.setKernel(cv.ml.SVM_LINEAR)
|
||||
svm.setTermCriteria((cv.TERM_CRITERIA_MAX_ITER, int(1e7), 1e-6))
|
||||
## [init]
|
||||
|
||||
#------------------------ 3. Train the svm ----------------------------------------------------
|
||||
## [train]
|
||||
svm.train(trainData, cv.ml.ROW_SAMPLE, labels)
|
||||
## [train]
|
||||
print('Finished training process')
|
||||
|
||||
#------------------------ 4. Show the decision regions ----------------------------------------
|
||||
## [show]
|
||||
green = (0,100,0)
|
||||
blue = (100,0,0)
|
||||
for i in range(I.shape[0]):
|
||||
for j in range(I.shape[1]):
|
||||
sampleMat = np.matrix([[j,i]], dtype=np.float32)
|
||||
response = svm.predict(sampleMat)[1]
|
||||
|
||||
if response == 1:
|
||||
I[i,j] = green
|
||||
elif response == 2:
|
||||
I[i,j] = blue
|
||||
## [show]
|
||||
|
||||
#----------------------- 5. Show the training data --------------------------------------------
|
||||
## [show_data]
|
||||
thick = -1
|
||||
# Class 1
|
||||
for i in range(NTRAINING_SAMPLES):
|
||||
px = trainData[i,0]
|
||||
py = trainData[i,1]
|
||||
cv.circle(I, (int(px), int(py)), 3, (0, 255, 0), thick)
|
||||
|
||||
# Class 2
|
||||
for i in range(NTRAINING_SAMPLES, 2*NTRAINING_SAMPLES):
|
||||
px = trainData[i,0]
|
||||
py = trainData[i,1]
|
||||
cv.circle(I, (int(px), int(py)), 3, (255, 0, 0), thick)
|
||||
## [show_data]
|
||||
|
||||
#------------------------- 6. Show support vectors --------------------------------------------
|
||||
## [show_vectors]
|
||||
thick = 2
|
||||
sv = svm.getUncompressedSupportVectors()
|
||||
|
||||
for i in range(sv.shape[0]):
|
||||
cv.circle(I, (int(sv[i,0]), int(sv[i,1])), 6, (128, 128, 128), thick)
|
||||
## [show_vectors]
|
||||
|
||||
cv.imwrite('result.png', I) # save the Image
|
||||
cv.imshow('SVM for Non-Linear Training Data', I) # show it to the user
|
||||
cv.waitKey()
|
||||
@@ -1,73 +0,0 @@
|
||||
#!/usr/bin/env python
|
||||
|
||||
import cv2 as cv
|
||||
import numpy as np
|
||||
|
||||
SZ=20
|
||||
bin_n = 16 # Number of bins
|
||||
|
||||
|
||||
affine_flags = cv.WARP_INVERSE_MAP|cv.INTER_LINEAR
|
||||
|
||||
## [deskew]
|
||||
def deskew(img):
|
||||
m = cv.moments(img)
|
||||
if abs(m['mu02']) < 1e-2:
|
||||
return img.copy()
|
||||
skew = m['mu11']/m['mu02']
|
||||
M = np.float32([[1, skew, -0.5*SZ*skew], [0, 1, 0]])
|
||||
img = cv.warpAffine(img,M,(SZ, SZ),flags=affine_flags)
|
||||
return img
|
||||
## [deskew]
|
||||
|
||||
## [hog]
|
||||
def hog(img):
|
||||
gx = cv.Sobel(img, cv.CV_32F, 1, 0)
|
||||
gy = cv.Sobel(img, cv.CV_32F, 0, 1)
|
||||
mag, ang = cv.cartToPolar(gx, gy)
|
||||
bins = np.int32(bin_n*ang/(2*np.pi)) # quantizing binvalues in (0...16)
|
||||
bin_cells = bins[:10,:10], bins[10:,:10], bins[:10,10:], bins[10:,10:]
|
||||
mag_cells = mag[:10,:10], mag[10:,:10], mag[:10,10:], mag[10:,10:]
|
||||
hists = [np.bincount(b.ravel(), m.ravel(), bin_n) for b, m in zip(bin_cells, mag_cells)]
|
||||
hist = np.hstack(hists) # hist is a 64 bit vector
|
||||
return hist
|
||||
## [hog]
|
||||
|
||||
img = cv.imread(cv.samples.findFile('digits.png'),0)
|
||||
if img is None:
|
||||
raise Exception("we need the digits.png image from samples/data here !")
|
||||
|
||||
|
||||
cells = [np.hsplit(row,100) for row in np.vsplit(img,50)]
|
||||
|
||||
# First half is trainData, remaining is testData
|
||||
train_cells = [ i[:50] for i in cells ]
|
||||
test_cells = [ i[50:] for i in cells]
|
||||
|
||||
###### Now training ########################
|
||||
|
||||
deskewed = [list(map(deskew,row)) for row in train_cells]
|
||||
hogdata = [list(map(hog,row)) for row in deskewed]
|
||||
trainData = np.float32(hogdata).reshape(-1,64)
|
||||
responses = np.repeat(np.arange(10),250)[:,np.newaxis]
|
||||
|
||||
svm = cv.ml.SVM_create()
|
||||
svm.setKernel(cv.ml.SVM_LINEAR)
|
||||
svm.setType(cv.ml.SVM_C_SVC)
|
||||
svm.setC(2.67)
|
||||
svm.setGamma(5.383)
|
||||
|
||||
svm.train(trainData, cv.ml.ROW_SAMPLE, responses)
|
||||
svm.save('svm_data.dat')
|
||||
|
||||
###### Now testing ########################
|
||||
|
||||
deskewed = [list(map(deskew,row)) for row in test_cells]
|
||||
hogdata = [list(map(hog,row)) for row in deskewed]
|
||||
testData = np.float32(hogdata).reshape(-1,bin_n*4)
|
||||
result = svm.predict(testData)[1]
|
||||
|
||||
####### Check Accuracy ########################
|
||||
mask = result==responses
|
||||
correct = np.count_nonzero(mask)
|
||||
print(correct*100.0/result.size)
|
||||
Reference in New Issue
Block a user