def get_features(directory):
features = []
for fn in iglob('%s/*.png' % directory):
image = color.rgb2gray(io.imread(fn))
features.append(get_histogram(image).reshape(-1))
features.append(get_histogram(np.fliplr(image)).reshape(-1))
return features
def get_detections(svm, target):
# Detection window size, must be same train image size
WIDTH, HEIGHT = (64, 64)
CELL_SIZE = 8
THRESHOLD = 3.0
target_scaled = target + 0
scale_factor = 2.0**(-1.0 / 8.0)
# Set result list
detections = []
for s in range(16):
# Get image histogram
histogram = get_histogram(target_scaled)
# Check detction window with sliding cell size
for y in range(0, histogram.shape[0] - int(HEIGHT / CELL_SIZE)):
for x in range(0, histogram.shape[1] - int(WIDTH / CELL_SIZE)):
# Get feature vector and score
feature = histogram[y:(y + int(HEIGHT / CELL_SIZE)),
x:(x + int(WIDTH / CELL_SIZE))].reshape(
1, -1)
# score = svm.decision_function(feature)
score = svm.predict(feature)
# if score[0] > THRESHOLD:
if score[0] == 1:
scale = (scale_factor**s)
detections.append({
'x': x * CELL_SIZE / scale,
'y': y * CELL_SIZE / scale,
'width': WIDTH / scale,
'height': HEIGHT / scale,
'score': score
})
target_scaled = transform.rescale(target_scaled, scale_factor)
return detections
def get_features(directory):
# 空配列を作成
features = []
# 変数 directory の中の PNG 形式の画像ファイル名を要素として取り出す
for fn in iglob('%s/*.png' % directory):
# グレースケールで画像を変数 image に格納
image = color.rgb2gray(io.imread(fn))
# get_histgram() 関数に画像データを渡し、画像一枚分のヒストグラムを取得し、一次元配列、つまり特徴ベクトルにする
# append で、要素として追加、一つの要素が配列なので注意、入れ子配列
features.append(get_histogram(image).reshape(-1))
# import pdb; pdb.set_trace()
# 学習データを増やすために fliplr() 関数で画像を反転させて、同じ処理を実施
features.append(get_histogram(np.fliplr(image)).reshape(-1))
# import pdb; pdb.set_trace()
return features
import sys
WIDTH, HEIGHT = (64, 64) # 検出窓サイズ、学習画像の大きさと同じ
CELL_SIZE = 4
THRESHOLD = 3.0
HEIGHT_SIZE = int(HEIGHT / CELL_SIZE)
WIDTH_SIZE = int(WIDTH / CELL_SIZE)
svm = pickle.load(open(sys.argv[1], 'rb'))
target = color.rgb2gray(io.imread(sys.argv[2]))
target_scaled = target + 0
scale_factor = 2.0**(-1.0 / 8.0)
detections = []
for s in range(16):
histogram = get_histogram(target_scaled)
for y in range(0, histogram.shape[0] - HEIGHT_SIZE):
for x in range(0, histogram.shape[1] - WIDTH_SIZE):
feature = histogram[y:y + HEIGHT_SIZE,
x:x + WIDTH_SIZE].reshape(-1)
score = svm.decision_function(feature)
if score[0] > THRESHOLD:
# 検出
scale = (scale_factor**s)
detections.append({
'x': x * CELL_SIZE / scale,
'y': y * CELL_SIZE / scale,
'width': WIDTH / scale,
'height': HEIGHT / scale,
import sys
WIDTH, HEIGHT = (64, 64) # 検出窓サイズ、学習画像の大きさと同じ
CELL_SIZE = 4
THRESHOLD = 3.0
HEIGHT_SIZE = int(HEIGHT / CELL_SIZE)
WIDTH_SIZE = int(WIDTH / CELL_SIZE)
svm = pickle.load(open(sys.argv[1], 'rb'))
target = color.rgb2gray(io.imread(sys.argv[2]))
target_scaled = target + 0
scale_factor = 2.0 ** (-1.0 / 8.0)
detections = []
for s in range(16):
histogram = get_histogram(target_scaled)
for y in range(0, histogram.shape[0] - HEIGHT_SIZE):
for x in range(0, histogram.shape[1] - WIDTH_SIZE):
feature = histogram[
y:y + HEIGHT_SIZE,
x:x + WIDTH_SIZE
].reshape(-1)
score = svm.decision_function(feature)
if score[0] > THRESHOLD:
# 検出
scale = (scale_factor ** s)
detections.append({
'x': x * CELL_SIZE / scale,
'y': y * CELL_SIZE / scale,
