def getMoreFeatures():
net = DecafNet()
features = []
labels = []
counter = 0
for participant in os.listdir(os.path.join(data_dir,image_dir)):
for sequence in os.listdir(os.path.join(data_dir,image_dir, participant)):
if sequence != ".DS_Store":
image_files = sorted(os.listdir(os.path.join(data_dir,image_dir, participant,sequence)))
cutoff = len(image_files)/2
image_files = image_files[cutoff::]
label_file = open(os.path.join(data_dir,label_dir, participant,sequence,image_files[-1][:-4]+"_emotion.txt"))
label = eval(label_file.read())
label_file.close()
for image_file in image_files:
print counter, image_file
imarray = numpy.asarray(Image.open(os.path.join(data_dir,image_dir, participant,sequence,image_file)))
scores = net.classify(imarray, center_only=True)
features.append(net.feature(feature_level))
labels.append(label)
counter += 1
numpy.save("featuresMore",numpy.array(features))
numpy.save("labelsMore",numpy.array(labels))
def getMoreFeatures():
net = DecafNet()
features = []
labels = []
counter = 0
for participant in os.listdir(os.path.join(data_dir, image_dir)):
for sequence in os.listdir(
os.path.join(data_dir, image_dir, participant)):
if sequence != ".DS_Store":
image_files = sorted(
os.listdir(
os.path.join(data_dir, image_dir, participant,
sequence)))
cutoff = len(image_files) / 2
image_files = image_files[cutoff::]
label_file = open(
os.path.join(data_dir, label_dir, participant, sequence,
image_files[-1][:-4] + "_emotion.txt"))
label = eval(label_file.read())
label_file.close()
for image_file in image_files:
print counter, image_file
imarray = numpy.asarray(
Image.open(
os.path.join(data_dir, image_dir, participant,
sequence, image_file)))
scores = net.classify(imarray, center_only=True)
features.append(net.feature(feature_level))
labels.append(label)
counter += 1
numpy.save("featuresMore", numpy.array(features))
numpy.save("labelsMore", numpy.array(labels))
def full_check_decaf(win_slide=5, win_size=1024, blob_name='fc6_cudanet_out'):
from decaf.scripts.imagenet import DecafNet
net = DecafNet()
clf = joblib.load("420_decaf/classifier_decaf.pkl")
g_raster = gdal.Open('20-21-22-part2.tif') # test.tif
# plt.axis('off')
# f, axarr = plt.subplots(n, n)
result = {}
cols = range(0, g_raster.RasterXSize - win_size, win_slide)
rows = range(0, g_raster.RasterYSize - win_size, win_slide)
full = len(rows) * len(cols)
count = 0
pbar = progressbar.ProgressBar(maxval=full).start()
for i in range(0, g_raster.RasterXSize - win_size, win_slide):
for j in range(0, g_raster.RasterYSize - win_size, win_slide):
img = get_sample(g_raster, i, j, win_size)
net.classify(img, True)
tmp = net.feature(blob_name) #与训练时候保持一致
result[(j,i)] = clf.predict(tmp)
if result[(j,i)] == 2:
io.imsave("420_decaf/slide_target/%s_%s_%s_%s.png" % (j, i, j+win_size, i+win_size), img)
pbar.update(count+1)
count = count + 1
pbar.finish()
arr = np.ones((len(rows), len(cols)))
for k, v in result.items():
if v != 0 and v[0] == 2:
arr[k[0]/win_slide, k[1]/win_slide] = v[0]
return arr
def getPeakFeatures():
net = DecafNet()
features = numpy.zeros((number_sequences, feature_length))
labels = numpy.zeros((number_sequences, 1))
counter = 0
# Maybe sort them
for participant in os.listdir(os.path.join(data_dir, image_dir)):
for sequence in os.listdir(
os.path.join(data_dir, image_dir, participant)):
if sequence != ".DS_Store":
image_files = sorted(
os.listdir(
os.path.join(data_dir, image_dir, participant,
sequence)))
image_file = image_files[-1]
print counter, image_file
imarray = cv2.imread(
os.path.join(data_dir, image_dir, participant, sequence,
image_file))
imarray = cv2.cvtColor(imarray, cv2.COLOR_BGR2GRAY)
scores = net.classify(imarray, center_only=True)
features[counter] = net.feature(feature_level) #.flatten()
label_file = open(
os.path.join(data_dir, label_dir, participant, sequence,
image_file[:-4] + "_emotion.txt"))
labels[counter] = eval(label_file.read())
label_file.close()
counter += 1
numpy.save("featuresPeak5", features)
numpy.save("labelsPeak5", labels)
def getPeakFaceFeatures():
net = DecafNet()
cascade = cv2.CascadeClassifier('haarcascade_frontalface_alt2.xml')
features = numpy.zeros((number_sequences,feature_length))
labels = numpy.zeros((number_sequences,1))
counter = 0
# Maybe sort them
for participant in os.listdir(os.path.join(data_dir,image_dir)):
for sequence in os.listdir(os.path.join(data_dir,image_dir, participant)):
if sequence != ".DS_Store":
image_files = sorted(os.listdir(os.path.join(data_dir,image_dir, participant,sequence)))
image_file = image_files[-1]
print counter, image_file
imarray = cv2.imread(os.path.join(data_dir,image_dir, participant,sequence,image_file))
imarray = cv2.cvtColor(imarray,cv2.COLOR_BGR2GRAY)
rects = cascade.detectMultiScale(imarray, 1.3, 3, cv2.cv.CV_HAAR_SCALE_IMAGE, (150,150))
if len(rects) > 0:
facerect=rects[0]
imarray = imarray[facerect[1]:facerect[1]+facerect[3], facerect[0]:facerect[0]+facerect[2]]
scores = net.classify(imarray, center_only=True)
features[counter] = net.feature(feature_level).flatten()
label_file = open(os.path.join(data_dir,label_dir, participant,sequence,image_file[:-4]+"_emotion.txt"))
labels[counter] = eval(label_file.read())
label_file.close()
counter += 1
numpy.save("featuresPeakFace5",features)
numpy.save("labelsPeakFace5",labels)
class AVIClassifier(object):
def __init__(self, decaf_folder = None, classifer_file = None):
if decaf_folder is None:
decaf_folder = '../models/imagenet_pretrained/'
if classifer_file is None:
classifer_file = "../models/lg_classifier_public"
self.net = DecafNet(path.join(decaf_folder, 'imagenet.decafnet.epoch90'),
path.join(decaf_folder, 'imagenet.decafnet.meta'))
self.feat_layer = 'fc6_cudanet_out'
self.classifier = cPickle.load(open(classifer_file, "r"))
def predict(self, img):
im = img_as_ubyte(color.rgb2gray(img))
scores = self.net.classify(im, center_only = True)
feats = self.net.feature(self.feat_layer).flatten()
defect_probs = self.classifier.predict_proba(feats)
return sorted(zip(self.classifier.classes_, defect_probs[0]), key = lambda (cls, prob): prob, reverse=True)
def class_names(self):
return self.classifier.classes_
def getPeakFaceFeatures():
net = DecafNet()
cascade = cv2.CascadeClassifier('haarcascade_frontalface_alt2.xml')
features = numpy.zeros((number_sequences, feature_length))
labels = numpy.zeros((number_sequences, 1))
counter = 0
# Maybe sort them
for participant in os.listdir(os.path.join(data_dir, image_dir)):
for sequence in os.listdir(
os.path.join(data_dir, image_dir, participant)):
if sequence != ".DS_Store":
image_files = sorted(
os.listdir(
os.path.join(data_dir, image_dir, participant,
sequence)))
image_file = image_files[-1]
print counter, image_file
imarray = cv2.imread(
os.path.join(data_dir, image_dir, participant, sequence,
image_file))
imarray = cv2.cvtColor(imarray, cv2.COLOR_BGR2GRAY)
rects = cascade.detectMultiScale(imarray, 1.3, 3,
cv2.cv.CV_HAAR_SCALE_IMAGE,
(150, 150))
if len(rects) > 0:
facerect = rects[0]
imarray = imarray[facerect[1]:facerect[1] + facerect[3],
facerect[0]:facerect[0] + facerect[2]]
scores = net.classify(imarray, center_only=True)
features[counter] = net.feature(feature_level).flatten()
label_file = open(
os.path.join(data_dir, label_dir, participant, sequence,
image_file[:-4] + "_emotion.txt"))
labels[counter] = eval(label_file.read())
label_file.close()
counter += 1
numpy.save("featuresPeakFace5", features)
numpy.save("labelsPeakFace5", labels)
class AVIClassifier(object):
def __init__(self, decaf_folder=None, classifer_file=None):
if decaf_folder is None:
decaf_folder = '../models/imagenet_pretrained/'
if classifer_file is None:
classifer_file = "../models/lg_classifier_public"
self.net = DecafNet(
path.join(decaf_folder, 'imagenet.decafnet.epoch90'),
path.join(decaf_folder, 'imagenet.decafnet.meta'))
self.feat_layer = 'fc6_cudanet_out'
self.classifier = cPickle.load(open(classifer_file, "r"))
def predict(self, img):
im = img_as_ubyte(color.rgb2gray(img))
scores = self.net.classify(im, center_only=True)
feats = self.net.feature(self.feat_layer).flatten()
defect_probs = self.classifier.predict_proba(feats)
return sorted(zip(self.classifier.classes_, defect_probs[0]),
key=lambda (cls, prob): prob,
reverse=True)
def class_names(self):
return self.classifier.classes_
def getPeakFeatures():
net = DecafNet()
features = numpy.zeros((number_sequences,feature_length))
labels = numpy.zeros((number_sequences,1))
counter = 0
# Maybe sort them
for participant in os.listdir(os.path.join(data_dir,image_dir)):
for sequence in os.listdir(os.path.join(data_dir,image_dir, participant)):
if sequence != ".DS_Store":
image_files = sorted(os.listdir(os.path.join(data_dir,image_dir, participant,sequence)))
image_file = image_files[-1]
print counter, image_file
imarray = cv2.imread(os.path.join(data_dir,image_dir, participant,sequence,image_file))
imarray = cv2.cvtColor(imarray,cv2.COLOR_BGR2GRAY)
scores = net.classify(imarray, center_only=True)
features[counter] = net.feature(feature_level)#.flatten()
label_file = open(os.path.join(data_dir,label_dir, participant,sequence,image_file[:-4]+"_emotion.txt"))
labels[counter] = eval(label_file.read())
label_file.close()
counter += 1
numpy.save("featuresPeak5",features)
numpy.save("labelsPeak5",labels)
class ConvNetFeatureExtractor(FeatureExtractor):
def __init__(
self,
feature_layer='fc7_cudanet_out',
pretrained_params='imagenet.decafnet.epoch90',
pretrained_meta='imagenet.decafnet.meta',
center_only=True
):
"""
:param feature_layer: The ConvNet layer that's used for
feature extraction. Defaults to
`fc7_cudanet_out`. A description of all
available layers for the
ImageNet-1k-pretrained ConvNet is found
in the DeCAF wiki. They are:
- `pool5_cudanet_out`
- `fc6_cudanet_out`
- `fc6_neuron_cudanet_out`
- `fc7_cudanet_out`
- `fc7_neuron_cudanet_out`
:param pretrained_params: This must point to the file with the
pretrained parameters. Defaults to
`imagenet.decafnet.epoch90`. For
the ImageNet-1k-pretrained ConvNet
this file can be obtained from here:
http://www.eecs.berkeley.edu/~jiayq/decaf_pretrained/
:param pretrained_meta: Similar to `pretrained_params`, this
must file to the file with the
pretrained parameters' metadata.
Defaults to `imagenet.decafnet.meta`.
:param center_only: Use the center patch of the image only
when extracting features. If `False`, use
four corners, the image center and flipped
variants and average a total of 10 feature
vectors, which will usually yield better
results. Defaults to `True`.
"""
super(ConvNetFeatureExtractor, self).__init__()
self.feature_layer = feature_layer
self.pretrained_params = pretrained_params
self.pretrained_meta = pretrained_meta
self.center_only = center_only
self.convnet = DecafNet(
self.pretrained_params,
self.pretrained_meta
)
def _extract(self, img):
"""
:param cv2 image
:return: np.array with shape (4096,)
"""
img = self.convnet.oversample(img, center_only=self.center_only)
self.convnet.classify_direct(img)
feat = self.convnet.feature(self.feature_layer)
if not self.center_only:
feat = feat.mean(0)
return feat[0]
class ConvNetFeatures(BaseEstimator):
"""Extract features from images using a pretrained ConvNet.
Based on Yangqing Jia and Jeff Donahue's `DeCAF
<https://github.com/UCB-ICSI-Vision-Group/decaf-release/wiki>`_.
Please make sure you read and accept DeCAF's license before you
use this class.
If ``classify_direct=False``, expects its input X to be a list of
image filenames or arrays as produced by
`np.array(Image.open(filename))`.
"""
verbose = 0
def __init__(
self,
feature_layer='fc7_cudanet_out',
pretrained_params='imagenet.decafnet.epoch90',
pretrained_meta='imagenet.decafnet.meta',
center_only=True,
classify_direct=False,
verbose=0,
):
"""
:param feature_layer: The ConvNet layer that's used for
feature extraction. Defaults to
`fc7_cudanet_out`. A description of all
available layers for the
ImageNet-1k-pretrained ConvNet is found
in the DeCAF wiki. They are:
- `pool5_cudanet_out`
- `fc6_cudanet_out`
- `fc6_neuron_cudanet_out`
- `fc7_cudanet_out`
- `fc7_neuron_cudanet_out`
- `probs_cudanet_out`
:param pretrained_params: This must point to the file with the
pretrained parameters. Defaults to
`imagenet.decafnet.epoch90`. For
the ImageNet-1k-pretrained ConvNet
this file can be obtained from here:
http://www.eecs.berkeley.edu/~jiayq/decaf_pretrained/
:param pretrained_meta: Similar to `pretrained_params`, this
must file to the file with the
pretrained parameters' metadata.
Defaults to `imagenet.decafnet.meta`.
:param center_only: Use the center patch of the image only
when extracting features. If `False`, use
four corners, the image center and flipped
variants and average a total of 10 feature
vectors, which will usually yield better
results. Defaults to `True`.
:param classify_direct: When `True`, assume that input X is an
array of shape (num x 256 x 256 x 3)
as returned by `prepare_image`.
"""
self.feature_layer = feature_layer
self.pretrained_params = pretrained_params
self.pretrained_meta = pretrained_meta
self.center_only = center_only
self.classify_direct = classify_direct
self.net_ = None
if (not os.path.exists(pretrained_params) or
not os.path.exists(pretrained_meta)):
raise ValueError(
"Pre-trained ConvNet parameters not found. You may"
"need to download the files from "
"http://www.eecs.berkeley.edu/~jiayq/decaf_pretrained/ and "
"pass the path to the two files as `pretrained_params` and "
"`pretrained_meta` to the `{}` estimator.".format(
self.__class__.__name__))
def fit(self, X=None, y=None):
from decaf.scripts.imagenet import DecafNet # soft dep
if self.net_ is None:
self.net_ = DecafNet(
self.pretrained_params,
self.pretrained_meta,
)
return self
@cache.cached(_transform_cache_key)
def transform(self, X):
features = []
for img in X:
if self.classify_direct:
images = self.net_.oversample(
img, center_only=self.center_only)
self.net_.classify_direct(images)
else:
if isinstance(img, str):
import Image # soft dep
img = np.array(Image.open(img))
self.net_.classify(img, center_only=self.center_only)
feat = None
for layer in self.feature_layer.split(','):
val = self.net_.feature(layer)
if feat is None:
feat = val
else:
feat = np.hstack([feat, val])
if not self.center_only:
feat = feat.flatten()
features.append(feat)
if self.verbose:
sys.stdout.write(
"\r[ConvNet] %d%%" % (100. * len(features) / len(X)))
sys.stdout.flush()
if self.verbose:
sys.stdout.write('\n')
return np.vstack(features)
def prepare_image(self, image):
"""Returns image of shape `(256, 256, 3)`, as expected by
`transform` when `classify_direct = True`.
"""
from decaf.util import transform # soft dep
_JEFFNET_FLIP = True
# first, extract the 256x256 center.
image = transform.scale_and_extract(transform.as_rgb(image), 256)
# convert to [0,255] float32
image = image.astype(np.float32) * 255.
if _JEFFNET_FLIP:
# Flip the image if necessary, maintaining the c_contiguous order
image = image[::-1, :].copy()
# subtract the mean
image -= self.net_._data_mean
return image
class DecafExtractor:
def __init__(
self,
layer_name,
model_path='dist/decaf-release/model/imagenet.decafnet.epoch90',
meta_path='dist/decaf-release/model/imagenet.decafnet.meta'):
self.layer_name = layer_name
self.net = DecafNet(model_path, meta_path)
self.transforms = [NopTransform()]
def set_parameters(self,
patch_size,
patches_per_image,
levels,
image_dim,
decaf_oversample=False):
self.patch_size = patch_size
self.patches_per_image = patches_per_image
self.levels = levels
self.image_dim = image_dim
self.decaf_oversample = decaf_oversample
self.patch_sizes = map(int,
self.patch_size * 2**np.arange(0, levels, 1.0))
def add_transform(self, transform):
self.transforms.append(transform)
def get_descriptor_size(self):
if self.layer_name in ['67_relu', '67']:
return 8192
else:
return 4096
def get_decaf(self, im):
scores = self.net.classify(np.asarray(im),
center_only=not self.decaf_oversample)
if self.layer_name == '67_relu':
return np.hstack([
self.net.feature('fc6_neuron_cudanet_out'),
self.net.feature('fc7_neuron_cudanet_out')
])
elif self.layer_name == '67':
return np.hstack([
self.net.feature('fc6_cudanet_out'),
self.net.feature('fc7_cudanet_out')
])
else:
return self.net.feature(self.layer_name)
def get_number_of_features_per_image(self):
if self.decaf_oversample:
return 10 * len(self.transforms)
else:
return len(self.transforms)
def extract_image(self, filename):
""" This method extracts 4096-dimensional DeCAF features from
patches at multiple scales belonging to the the image file <filename>.
Number of scales, patch size, and other settings are set by method
set_parameters().
This method returns tuple (patches, positions, patch_number),
where <patches> is a numpy array of dimension (patch_number, 4096) holding features,
<positions> is a numpy array of dimension (patch_number, 2) holding absolute positions of patches,
<patch_number> is the number of patches that algorithm managed to extract -- it
is guaranteed to be at most the number of originally specified.
"""
log = get_logger()
im = Image.open(filename)
# Resizing image
if max(im.size) != self.image_dim:
if im.size[0] > im.size[1]:
new_height = (self.image_dim * im.size[1]) / im.size[0]
new_dim = (self.image_dim, new_height)
else:
new_width = (self.image_dim * im.size[0]) / im.size[1]
new_dim = (new_width, self.image_dim)
log.info('Resizing image from (%d, %d) to (%d, %d).', im.size[0],
im.size[1], new_dim[0], new_dim[1])
im = im.resize(new_dim, Image.ANTIALIAS)
# Estimating number of extracted features taking into account transformations
estimated_feature_num = self.patches_per_image * self.get_number_of_features_per_image(
)
# Provisioning space for patches and locations
feature_storage = ExtractedFeatures(estimated_feature_num,
self.get_descriptor_size())
log.info(
'Extracting up to %d patches at %d levels from "%s"...',
self.patches_per_image * self.get_number_of_features_per_image(),
self.levels, basename(filename))
# Applying transformations and extracting features
for xform in self.transforms:
im_ = xform.apply(im)
self.extract(im_, feature_storage, xform.check_coords, xform,
filename)
log.info('Done. Extracted: %d.', feature_storage.cursor)
return feature_storage
def extract(self, im, feature_storage, check_patch_coords, transform,
filename):
(w, h) = im.size
# Calculating patch step
if self.levels > 0:
patch_step = int(
(w * h * len(self.patch_sizes) / self.patches_per_image)**0.5)
w_steps = np.arange(0, w, patch_step)
h_steps = np.arange(0, h, patch_step)
(xx, yy) = np.meshgrid(w_steps, h_steps)
if isinstance(transform, NopTransform): # Hacky....
# Extracting features for the whole image
feature_storage.append(self.get_decaf(im), np.matrix([0, 0]))
# Extracting features from patches
for l in range(self.levels):
for i in range(xx.shape[0]):
for j in range(xx.shape[1]):
x = xx[i, j]
y = yy[i, j]
patch_left = x + self.patch_sizes[l]
patch_bottom = y + self.patch_sizes[l]
if (check_patch_coords(x, y, patch_left, patch_bottom)
and patch_left <= w and patch_bottom <= h):
patch = im.crop((x, y, patch_left, patch_bottom))
patch.load()
feature_storage.append(self.get_decaf(patch),
np.matrix([x, y]))
class NetworkDecaf(Network):
"""
Implementation for the Decaf library.
"""
def __init__(self, model_spec_filename, model_filename=None,\
wnid_words_filename=None, center_only=False, wnid_subset = []):
"""
*** PRIVATE CONSTRUCTOR ***
"""
# the following is just an hack to allow retro-compatibility
# with existing code
if isinstance(model_spec_filename, NetworkDecafParams):
params = model_spec_filename
model_spec_filename = params.model_spec_filename
model_filename = params.model_filename
wnid_words_filename = params.wnid_words_filename
center_only = params.center_only
wnid_subset = params.wnid_subset
if wnid_subset != []:
print 'Warning: subset of labels not supported yet'
else:
assert isinstance(model_spec_filename, str)
assert model_filename != None
assert wnid_words_filename != None
# load Decaf model
self.net_ = DecafNet(model_filename, model_spec_filename)
self.center_only_ = center_only
# build a dictionary label --> description
self.dict_label_desc_ = {}
dict_desc_label = {}
fd = open(wnid_words_filename)
for line in fd:
temp = line.strip().split('\t')
wnid = temp[1].strip()
self.dict_label_desc_[wnid] = temp[2].strip()
dict_desc_label[temp[2].split(',')[0]] = wnid
fd.close()
# build a dictionary label --> label_id
self.dict_label_id_ = {}
self.labels_ = []
for i, desc in enumerate(self.net_.label_names):
self.dict_label_id_[dict_desc_label[desc]] = i
self.labels_.append(dict_desc_label[desc])
# Load the mean vector from file
# mean of 3 channels
self.net_.mean_img = np.mean(np.mean(self.net_._data_mean, axis=1),
axis=0)
# it is in BGR convert in RGB
#self.net_.mean_img = self.net_.mean_img[::-1]
def get_mean_img(self):
return self.net_.mean_img
def get_input_dim(self):
return decaf.scripts.imagenet.INPUT_DIM
def get_label_id(self, label):
return self.dict_label_id_[label]
def get_label_desc(self, label):
return self.dict_label_desc_[label]
def get_labels(self):
return self.labels_
def evaluate(self, img, layer_name='softmax'):
# for now only center_only is supported
assert self.center_only_ == True
# first, extract the 227x227 center
dim = decaf.scripts.imagenet.INPUT_DIM
image = util.crop_image_center(decaf.util.transform.as_rgb(img))
image = skimage.transform.resize(image, (dim, dim))
# convert to [0,255] float32
image = image.astype(np.float32) * 255.
assert np.max(image) <= 255
# Flip the image if necessary, maintaining the c_contiguous order
if decaf.scripts.imagenet._JEFFNET_FLIP:
image = image[::-1, :].copy()
# subtract the mean, cropping the 256x256 mean image
xoff = (self.net_._data_mean.shape[1] - dim) / 2
yoff = (self.net_._data_mean.shape[0] - dim) / 2
image -= self.net_._data_mean[yoff + yoff + dim, xoff:xoff + dim]
# make sure the data in contiguous in memory
images = np.ascontiguousarray(image[np.newaxis], dtype=np.float32)
# classify
predictions = self.net_.classify_direct(images)
scores = predictions.mean(0)
# look at the particular layer
if layer_name == 'softmax':
return scores
elif layer_name == 'fc7_relu':
layer_name = 'fc7_neuron_cudanet_out'
elif layer_name == 'fc7':
layer_name = 'fc7_cudanet_out'
elif layer_name == 'fc6_relu':
layer_name = 'fc6_neuron_cudanet_out'
elif layer_name == 'fc6':
layer_name = 'fc6_cudanet_out'
elif layer_name == 'pool5':
layer_name = 'pool5_cudanet_out'
else:
raise ValueError('layer_name not supported')
return self.net_.feature(layer_name)
print 'Total region count:', numFeatures
#test
img = None
TEST = False
if TEST == True:
feature = layer.GetNextFeature()
img = getRegion(g_raster, feature)
# show
#print img.dtype
#plt.figure()
#plt.imshow(img)
net.classify(img, True)
tmp = net.feature(blob_name) #与训练时候保持一致
is_slide = clf.predict(tmp)
feature.SetField("slide", is_slide[0])
else:
# loop through the regions and predict them
pbar = progressbar.ProgressBar(maxval=numFeatures).start()
cnt = 0
feature = layer.GetNextFeature()
while feature:
# 获取对应的图像样本
img = getRegion(g_raster, feature)
#imshow(img)
#raw_input()
return resized, np.asarray(resized)
if __name__ == '__main__':
folder = sys.argv[1]
out_folder = sys.argv[2]
if os.path.exists(out_folder):
shutil.rmtree(out_folder)
os.mkdir(out_folder)
imgs = glob.glob(os.path.join(folder,"*.jpeg"))
print imgs
net = DecafNet('../imagenet_pretrained/imagenet.decafnet.epoch90', '../imagenet_pretrained/imagenet.decafnet.meta')
flog = open('log.txt', 'w')
for i, imgname in enumerate(imgs):
flog.write("%s\t%d" % (imgname, i))
try:
resized, img = load_and_resize(imgname)
except ValueError:
print "error when read %s" % imgname
continue
scores = net.classify(img, center_only=True)
feature = net.feature('fc6_cudanet_out')
print feature
out_file = open(os.path.join(out_folder, "%d.npy" % i), 'w')
np.save(out_file, feature)
io.imsave(os.path.join(out_folder, "%d.jpg" % i), resized)
class DecafExtractor:
def __init__(self, layer_name,
model_path = 'dist/decaf-release/model/imagenet.decafnet.epoch90',
meta_path = 'dist/decaf-release/model/imagenet.decafnet.meta'):
self.layer_name = layer_name
self.net = DecafNet(model_path, meta_path)
self.transforms = [NopTransform()]
def set_parameters(self, patch_size, patches_per_image, levels, image_dim, decaf_oversample=False):
self.patch_size = patch_size
self.patches_per_image = patches_per_image
self.levels = levels
self.image_dim = image_dim
self.decaf_oversample = decaf_oversample
self.patch_sizes = map(int, self.patch_size * 2**np.arange(0,levels,1.0))
def add_transform(self, transform):
self.transforms.append(transform)
def get_descriptor_size(self):
if self.layer_name in ['67_relu', '67']:
return 8192
else:
return 4096
def get_decaf(self, im):
scores = self.net.classify(np.asarray(im), center_only=not self.decaf_oversample)
if self.layer_name == '67_relu':
return np.hstack([self.net.feature('fc6_neuron_cudanet_out'),
self.net.feature('fc7_neuron_cudanet_out')])
elif self.layer_name == '67':
return np.hstack([self.net.feature('fc6_cudanet_out'),
self.net.feature('fc7_cudanet_out')])
else:
return self.net.feature(self.layer_name)
def get_number_of_features_per_image(self):
if self.decaf_oversample:
return 10*len(self.transforms)
else:
return len(self.transforms)
def extract_image(self, filename):
""" This method extracts 4096-dimensional DeCAF features from
patches at multiple scales belonging to the the image file <filename>.
Number of scales, patch size, and other settings are set by method
set_parameters().
This method returns tuple (patches, positions, patch_number),
where <patches> is a numpy array of dimension (patch_number, 4096) holding features,
<positions> is a numpy array of dimension (patch_number, 2) holding absolute positions of patches,
<patch_number> is the number of patches that algorithm managed to extract -- it
is guaranteed to be at most the number of originally specified.
"""
log = get_logger()
im = Image.open(filename)
# Resizing image
if max(im.size) != self.image_dim:
if im.size[0] > im.size[1]:
new_height = (self.image_dim * im.size[1]) / im.size[0]
new_dim = (self.image_dim, new_height)
else:
new_width = (self.image_dim * im.size[0]) / im.size[1]
new_dim = (new_width, self.image_dim)
log.info('Resizing image from (%d, %d) to (%d, %d).', im.size[0], im.size[1], new_dim[0], new_dim[1])
im = im.resize(new_dim, Image.ANTIALIAS)
# Estimating number of extracted features taking into account transformations
estimated_feature_num = self.patches_per_image * self.get_number_of_features_per_image()
# Provisioning space for patches and locations
feature_storage = ExtractedFeatures(estimated_feature_num, self.get_descriptor_size())
log.info('Extracting up to %d patches at %d levels from "%s"...',
self.patches_per_image * self.get_number_of_features_per_image(),
self.levels, basename(filename))
# Applying transformations and extracting features
for xform in self.transforms:
im_ = xform.apply(im)
self.extract(im_, feature_storage, xform.check_coords, xform, filename)
log.info('Done. Extracted: %d.', feature_storage.cursor)
return feature_storage
def extract(self, im, feature_storage, check_patch_coords, transform, filename):
(w, h) = im.size
# Calculating patch step
if self.levels > 0:
patch_step = int( (w*h * len(self.patch_sizes) / self.patches_per_image)**0.5 )
w_steps = np.arange(0, w, patch_step)
h_steps = np.arange(0, h, patch_step)
(xx, yy) = np.meshgrid(w_steps, h_steps)
if isinstance(transform, NopTransform): # Hacky....
# Extracting features for the whole image
feature_storage.append( self.get_decaf(im), np.matrix([0,0]) )
# Extracting features from patches
for l in range(self.levels):
for i in range(xx.shape[0]):
for j in range(xx.shape[1]):
x = xx[i,j]
y = yy[i,j]
patch_left = x+self.patch_sizes[l]
patch_bottom = y+self.patch_sizes[l]
if (check_patch_coords(x, y, patch_left, patch_bottom) and
patch_left <= w and patch_bottom <= h ):
patch = im.crop( (x, y, patch_left, patch_bottom) )
patch.load()
feature_storage.append( self.get_decaf(patch), np.matrix([x, y]) )
from skimage import io
import numpy as np
import scipy.io as sio
import os
import sys
if len(sys.argv) != 5:
print "Usage ", sys.argv[0], "<model_root_dir> <image_dir> <output_feature_path> <num_imgs>"
exit(1)
model_root = sys.argv[1]
net = DecafNet(model_root + 'imagenet.decafnet.epoch90', model_root + 'imagenet.decafnet.meta')
img_dir = sys.argv[2]
feature_path = sys.argv[3]
NUM_IMGS = int(sys.argv[4])
FEATURE_DIM = 4096 #fc6_cudanet_out's dimension
features = np.zeros((NUM_IMGS,FEATURE_DIM))
for i in range(NUM_IMGS):
filename = img_dir + "/%05d.jpg" %(i+1)
if os.path.exists(filename):
sys.stdout.write("Extracting DeCAF feature from image %d\n" %(i+1))
img = io.imread(filename)
net.classify(img, center_only=True)
features[i,:] = net.feature('fc6_cudanet_out')
sio.savemat(feature_path,{'features':features})
class NetworkDecaf(Network):
"""
Implementation for the Decaf library.
"""
def __init__(self, model_spec_filename, model_filename=None,\
wnid_words_filename=None, center_only=False, wnid_subset = []):
"""
*** PRIVATE CONSTRUCTOR ***
"""
# the following is just an hack to allow retro-compatibility
# with existing code
if isinstance(model_spec_filename, NetworkDecafParams):
params = model_spec_filename
model_spec_filename = params.model_spec_filename
model_filename = params.model_filename
wnid_words_filename = params.wnid_words_filename
center_only = params.center_only
wnid_subset = params.wnid_subset
if wnid_subset!=[]:
print 'Warning: subset of labels not supported yet'
else:
assert isinstance(model_spec_filename, str)
assert model_filename != None
assert wnid_words_filename != None
# load Decaf model
self.net_ = DecafNet(model_filename, model_spec_filename)
self.center_only_ = center_only
# build a dictionary label --> description
self.dict_label_desc_ = {}
dict_desc_label = {}
fd = open(wnid_words_filename)
for line in fd:
temp = line.strip().split('\t')
wnid = temp[1].strip()
self.dict_label_desc_[wnid] = temp[2].strip()
dict_desc_label[temp[2].split(',')[0]] = wnid
fd.close()
# build a dictionary label --> label_id
self.dict_label_id_ = {}
self.labels_ = []
for i, desc in enumerate(self.net_.label_names):
self.dict_label_id_[dict_desc_label[desc]] = i
self.labels_.append(dict_desc_label[desc])
# Load the mean vector from file
# mean of 3 channels
self.net_.mean_img =np.mean(np.mean(self.net_._data_mean,axis=1),axis=0)
# it is in BGR convert in RGB
#self.net_.mean_img = self.net_.mean_img[::-1]
def get_mean_img(self):
return self.net_.mean_img
def get_input_dim(self):
return decaf.scripts.imagenet.INPUT_DIM
def get_label_id(self, label):
return self.dict_label_id_[label]
def get_label_desc(self, label):
return self.dict_label_desc_[label]
def get_labels(self):
return self.labels_
def evaluate(self, img, layer_name = 'softmax'):
# for now only center_only is supported
assert self.center_only_ == True
# first, extract the 227x227 center
dim = decaf.scripts.imagenet.INPUT_DIM
image = util.crop_image_center(decaf.util.transform.as_rgb(img))
image = skimage.transform.resize(image, (dim, dim))
# convert to [0,255] float32
image = image.astype(np.float32) * 255.
assert np.max(image) <= 255
# Flip the image if necessary, maintaining the c_contiguous order
if decaf.scripts.imagenet._JEFFNET_FLIP:
image = image[::-1, :].copy()
# subtract the mean, cropping the 256x256 mean image
xoff = (self.net_._data_mean.shape[1] - dim)/2
yoff = (self.net_._data_mean.shape[0] - dim)/2
image -= self.net_._data_mean[yoff+yoff+dim, xoff:xoff+dim]
# make sure the data in contiguous in memory
images = np.ascontiguousarray(image[np.newaxis], dtype=np.float32)
# classify
predictions = self.net_.classify_direct(images)
scores = predictions.mean(0)
# look at the particular layer
if layer_name == 'softmax':
return scores
elif layer_name == 'fc7_relu':
layer_name = 'fc7_neuron_cudanet_out'
elif layer_name == 'fc7':
layer_name = 'fc7_cudanet_out'
elif layer_name == 'fc6_relu':
layer_name = 'fc6_neuron_cudanet_out'
elif layer_name == 'fc6':
layer_name = 'fc6_cudanet_out'
elif layer_name == 'pool5':
layer_name = 'pool5_cudanet_out'
else:
raise ValueError('layer_name not supported')
return self.net_.feature(layer_name)
