1

我需要使用增强级联训练对 scikit-learn 中的一些图像进行分类。我想根据HoG特征进行分类。

我下面的代码改编自这个例子。

这部分代码是我唯一真正做过的事情：

import sys
from scipy import misc, ndimage
from skimage import data, io, filter, color, exposure
from skimage.viewer import ImageViewer
from skimage.feature import hog
from skimage.transform import resize
import matplotlib.pyplot as plt
from sklearn.datasets import make_gaussian_quantiles
from sklearn.ensemble import AdaBoostClassifier
from sklearn.externals.six.moves import xrange
from sklearn.metrics import accuracy_score
from sklearn.tree import DecisionTreeClassifier
import pylab as pl
from sklearn.externals.six.moves import zip

f = open("PATH_TO_SAMPLES_LIST\\samples.txt",'r')
out = f.read().splitlines()
import numpy as np

### THIS IS THE MAIN CHANGES I MADE TO THE CODE
### THE CHANGES ARE ONLY IN ORDER TO GET HOG FEATURES OUT OF IMAGES TO PASS ON TO THE CLASSIFIERS

imgs = []
tmp_hogs = []

# I've omitted the code where I populate an array called "labels", but it's just a 1D #array with 528 elements, either 1 or 0

i=0
for file in out:
        filepath = "C:\\work_asaaki\\caltech\\cars_brad\\resized\\"
        readfile = filepath + file
        curr_img = color.rgb2gray(io.imread(readfile))
        imgs.append(curr_img)
        fd, hog_image = hog(curr_img, orientations=8, pixels_per_cell=(8, 8),
                 cells_per_block=(1, 1), visualise=True, normalise=True)
        tmp_hogs.append(fd) 
        i+=1

img_hogs = np.array(tmp_hogs, dtype =float) 
print img_hogs.shape

n_split = 508
X_train, X_test = np.array(img_hogs[:n_split]), np.array(img_hogs[n_split:])
y_train, y_test = np.array(labels[:n_split]), np.array(labels[n_split:])

其余代码来自链接上的示例：

#### THE CODE BELOW IS TAKEN DIRECTLY FROM THE EXAMPLE


bdt_real = AdaBoostClassifier(
    DecisionTreeClassifier(max_depth=2),
    n_estimators=600,
    learning_rate=1)

bdt_discrete = AdaBoostClassifier(
    DecisionTreeClassifier(max_depth=2),
    n_estimators=600,
    learning_rate=1.5,
    algorithm="SAMME")

bdt_real.fit(X_train, y_train)
bdt_discrete.fit(X_train, y_train)

real_test_errors = []
discrete_test_errors = []

for real_test_predict, discrete_train_predict in zip(
        bdt_real.staged_predict(X_test), bdt_discrete.staged_predict(X_test)):
    real_test_errors.append(
        1. - accuracy_score(real_test_predict, y_test))
    discrete_test_errors.append(
        1. - accuracy_score(discrete_train_predict, y_test))

n_trees_discrete = len(bdt_discrete)
n_trees_real = len(bdt_real)

# Boosting might terminate early but the following arrays are always
# n_estimators long. We crop them to the actual number of tree here:
discrete_estimator_errors = bdt_discrete.estimator_errors_[:n_trees_discrete]
real_estimator_errors = bdt_real.estimator_errors_[:n_trees_real]
discrete_estimator_weights = bdt_discrete.estimator_weights_[:n_trees_discrete]

plt.figure(figsize=(15, 5))

plt.subplot(131)
plt.plot(xrange(1, n_trees_discrete + 1),
         discrete_test_errors, c='black', label='SAMME')
plt.plot(xrange(1, n_trees_real + 1),
         real_test_errors, c='black',
         linestyle='dashed', label='SAMME.R')
plt.legend()
plt.ylim(0.18, 0.62)
plt.ylabel('Test Error')
plt.xlabel('Number of Trees')
print "n trees"
print n_trees_discrete
print "discrete_test_errors"
print bdt_discrete.estimator_errors_.shape

plt.subplot(132)
plt.plot(xrange(1, n_trees_discrete + 1), discrete_estimator_errors,
         "b", label='SAMME', alpha=.5)
plt.plot(xrange(1, n_trees_real + 1), real_estimator_errors,
         "r", label='SAMME.R', alpha=.5)
plt.legend()
plt.ylabel('Error')
plt.xlabel('Number of Trees')
plt.ylim((.2,
         max(real_estimator_errors.max(),
             discrete_estimator_errors.max()) * 1.2))
plt.xlim((-20, len(bdt_discrete) + 20))

print  "plotting..."

plt.subplot(133)
plt.plot(xrange(1, n_trees_discrete + 1), discrete_estimator_weights,
         "b", label='SAMME')
plt.legend()
plt.ylabel('Weight')
plt.xlabel('Number of Trees')
plt.ylim((0, discrete_estimator_weights.max() * 1.2))
plt.xlim((-20, n_trees_discrete + 20))
# prevent overlapping y-axis labels
plt.subplots_adjust(wspace=0.25)
plt.show()

我的问题是，这是根据 HoG 特征对图像进行分类的正确方法吗？共有 528 张图片。首先它们是 240x360。但是当我打印 的形状时img_hogs，我得到了：

(528L, 2640L)

有人告诉我没有要绘制的图表，因为分类会提前终止，因为特征比图像多得多。所以我将图像缩小到 20x30。

现在，当我打印 的形状时img_hogs，我得到：

(528L, 48L)

但我仍然没有得到任何结果。在任何一种情况下，都绘制了轴，但图形为空。

4

1 回答 1

1

我会删除这个问题，但我突然想到另一个分类新手可能有同样的问题，所以我只是回答这个问题。

问题是因为一个非常愚蠢的错误——我的训练数组labels没有负数，它完全由一个类的标签组成。

于 2014-04-20T14:48:12.373 回答