def extract_from_indices_file():
base_output_dir = (
"/mnt/data/tiny_images/py-tiny-image-access/loaded_images/cifar100"
)
# the first 50'000 indices are for training
train_dir = base_output_dir + "/train"
# the last 10'000 indices are for testing
test_dir = base_output_dir + "/test"
tinyimage.openTinyImage()
cifar100_indices = get_indices()
for i, index in enumerate(progressbar(cifar100_indices)):
if i < 50000:
output_dir = train_dir
else:
output_dir = test_dir
meta = tinyimage.getMetaData(index)
tinyimage.sliceToImage(tinyimage.sliceToBin(index), output_dir + "/" + meta[1])
tinyimage.closeTinyImage()
means = [] convergence = 10 max_iterations = 5 #get the samples samples = [] print "Loading images..." tinyimage.openTinyImage() for keyword in keywords: indexes = tinyimage.retrieveByTerm(keyword, max_pics) temp = [] for indx in indexes: samples.append(tinyimage.sliceToBin(indx).reshape(32,32,3, order="F")) tinyimage.closeTinyImage() print str(len(samples)) + " samples loaded..." print "Seeding..." #select random seeds from the samples for i in range(0, cluster_count): r = random.randint(0, max_pics * cluster_count) means.append(numpy.empty(cluster_dimension)) means[i][:] = samples[r] #perform k means print "Clustering..." a1 = assignclusters(means, samples)
import sys
sys.path.append("/home/charles/PyTinyImage")
import tinyimage as tinyimage
import leargist
import scipy
import struct
start = int(sys.argv[1])
stop = int(sys.argv[2])
f = open('/mnt/ti-%i.gists' % start, "w+")
tinyimage.openTinyImage()
for i in xrange(start, stop):
s = tinyimage.sliceToBin(i)
image = scipy.misc.toimage(s.reshape(32,32,3, order="F").copy())
gist = leargist.color_gist(image)
print len(gist)
f.write(struct.pack('f' * 960, *gist))
print "Remaining: %i" % (stop - i)
tinyimage.closeTinyImage()
f.close()
for label in f:
labels.append(label.rstrip())
return labels
def get_indices():
indices = []
with open('./indices_cifar100','r') as f:
for index in f:
indices.append(int(index.rstrip()))
return indices
if __name__ == "__main__":
keywords = get_labels()
tinyimage.openTinyImage()
images = []
ignore_indices = get_indices()
pick = len(ignore_indices)
for keyword in progressbar(keywords):
indexes = tinyimage.retrieveByTerm(keyword)
for i in indexes:
if i not in ignore_indices:
image = tinyimage.sliceToBin(i).reshape(32,32,3, order="F").astype('float32') / 255.
images.append(image)
relevant = np.array(images)
np.random.shuffle(relevant)
relevant = relevant[:pick]
np.save("relevant_images",relevant)
tinyimage.closeTinyImage()
import sys
import os
import tinyimage
if (len(sys.argv) < 4):
print "usage: python loadFromKeywords <keyword,keyword,keyword,...> <max> <output path>"
sys.exit(1)
#todo - start here
keywords = sys.argv[1].split(',')
max_pics = int(sys.argv[2])
output_path = sys.argv[3]
tinyimage.openTinyImage()
for keyword in keywords:
os.mkdir(output_path+"/"+keyword)
indexes = tinyimage.retrieveByTerm(keyword, max_pics)
for i in indexes:
tinyimage.sliceToImage(tinyimage.sliceToBin(i), output_path+"/"+keyword +"/"+str(i)+".png")
tinyimage.closeTinyImage()
import sys
import os
import tinyimage
if (len(sys.argv) < 4):
print "usage: python loadFromKeywords <keyword,keyword,keyword,...> <max> <output path>"
sys.exit(1)
#todo - start here
keywords = sys.argv[1].split(',')
max_pics = int(sys.argv[2])
output_path = sys.argv[3]
tinyimage.openTinyImage()
for keyword in keywords:
os.mkdir(output_path + "/" + keyword)
indexes = tinyimage.retrieveByTerm(keyword, max_pics)
for i in indexes:
tinyimage.sliceToImage(
tinyimage.sliceToBin(i),
output_path + "/" + keyword + "/" + str(i) + ".png")
tinyimage.closeTinyImage()
import sys
import os
import tinyimage
if len(sys.argv) < 4:
print(
"usage: python loadFromKeywords <keyword,keyword,keyword,...> <max> <output path>"
)
sys.exit(1)
keywords = sys.argv[1].split(",")
max_pics = int(sys.argv[2])
output_path = sys.argv[3]
tinyimage.openTinyImage()
for keyword in keywords:
keyword_output_dir = output_path + "/" + keyword
if not os.path.exists(keyword_output_dir):
os.mkdir(keyword_output_dir)
indexes = tinyimage.retrieveByTerm(keyword, max_pics)
for i in indexes:
tinyimage.sliceToImage(tinyimage.sliceToBin(i),
keyword_output_dir + "/" + str(i) + ".png")
tinyimage.closeTinyImage()
