jacobgil / pyfishervector

Python implementation for Image Classification based on GMM dictionaries and fisher vectors.

pyfishervector's Issues

Questions about Fisher vector

When I get the vector， I find some values is ‘Nan’ . What's going on?

ValueError: Array 'mean' must be a vector of length 16384.

I have tried a few hours to solve this problem,but i failed.almost in line 38, the likelihood_statistics function,g = [multivariate_normal(mean=means[k], cov=covs[k]) for k in range(0, len(weights))],i think this is the main problem,but i can't solve it,can you help me?thank you very much.

ValueError: The number of classes has to be greater than one; got 1

Hello,
When try to run the code,I meet the error like followings:
('Calculating descriptos. Number of images is', 5)
('Training GMM of size', 5)
(5, 514)
(5,)
Traceback (most recent call last):
File "pyFisherVector.py", line 146, in
classifier = train(gmm, fisher_features)
File "pyFisherVector.py", line 116, in train
clf.fit(X, Y)
File "/usr/local/lib/python2.7/dist-packages/sklearn/svm/base.py", line 150, in fit
y = self._validate_targets(y)
File "/usr/local/lib/python2.7/dist-packages/sklearn/svm/base.py", line 506, in _validate_targets
% len(cls))
ValueError: The number of classes has to be greater than one; got 1

I just put 5 images in a folder,I have printed Y is (0,0,0,0,0).Is it the reason why show the error that the number of classes just got 1?If you know how to solve this problem,could you share with me?Thanks!

Multivariate Normal Issue

I am using the latest version of opencv and several functions have been modified. I adapted the code to work with the current version, however, when function g = [multivariate_normal(mean=means[k], cov=covs[k]) for k in range(0, len(weights)) ] is applied the following error is returned:numpy.linalg.linalg.LinAlgError: singular matrix.

I know that some function is returning a singular matrix, which makes it impossible to apply the function multivariate_normal.

What should I do? I'm really needing FisherVector applied to the Bag Of Words, and that was the only implementation I found.

Sorry for the bad english and thanks for the help.

What is 'pdf' in line 38 'gaussians[index] = np.array([g_k.pdf(x) for g_k in g])' ?

What is 'pdf' in line 38 'gaussians[index] = np.array([g_k.pdf(x) for g_k in g])' ?????
I can't find the function @hendrycks @hugos94 @ @jacobgil

TypeError: The first input argument needs to be a sequence

hi, when my program run into
def train(gmm,features):
X = np.concatenate(features.values())
Y = np.concatenate([np.float32([i]*len(v)) for i,v in zip(range(0, len(features)), features.values())])
clf = svm.SVC()
clf.fit(X,Y)
return clf
report the error :
in train
X = np.concatenate(features.values())
TypeError: The first input argument needs to be a sequence

can you help me resolve this? Thanks

ValueError: need at least one array to concatenate

def image_descriptors(file):
img = cv2.imread(file, 0)
img = cv2.resize(img, (256, 256))
_ , descriptors = cv2.SIFT().detectAndCompute(img, None)
return descriptors

def folder_descriptors(folder):
files = glob.glob(folder + "/*.jpg")
print("Calculating descriptos. Number of images is", len(files))
return np.concatenate([image_descriptors(file) for file in files])

this error is showing at last line i.e.
return np.concatenate([image_descriptors(file) for file in files])

how to use this code?

@jacobgil hi
Sorry for raising this issue.But when I tried to use fisher.py for cluster my images by using this command:

python fisher.py -d ./images

An error occupied.The log is here:

('Calculating descriptos. Number of images is', 0)
Traceback (most recent call last):
File "fisher.py", line 129, in
gmm = load_gmm(working_folder) if args.loadgmm else generate_gmm(working_folder, args.number)
File "fisher.py", line 76, in generate_gmm
words = np.concatenate([folder_descriptors(folder) for folder in glob.glob(input_folder + '/*')])
File "fisher.py", line 26, in folder_descriptors
return np.concatenate([image_descriptors(file) for file in files])
ValueError: need at least one array to concatenate

The "./images" folder just included all the images in it and has no sub folders :
./images
--img1.jpg
--img2.jpg
...

Could you please give some help?
Thank you !

ValueError: zero-dimensional arrays cannot be concatenated

Hi,

I'm trying to run the code but I receive the following error:

Traceback (most recent call last):
  File "C:\Users\dangz\Documents\Fotolog\Aesthetic\descriptors\fish.py", line 139, in <module>
    fisher_features = fisher_features(working_folder, gmm)
  File "C:\Users\dangz\Documents\Fotolog\Aesthetic\descriptors\fish.py", line 102, in fisher_features
    features = {f : get_fisher_vectors_from_folder(f, gmm) for f in folders}
  File "C:\Users\dangz\Documents\Fotolog\Aesthetic\descriptors\fish.py", line 102, in <dictcomp>
    features = {f : get_fisher_vectors_from_folder(f, gmm) for f in folders}
  File "C:\Users\dangz\Documents\Fotolog\Aesthetic\descriptors\fish.py", line 97, in get_fisher_vectors_from_folder
    return np.float32([fisher_vector(image_descriptors(file), *gmm) for file in files])
  File "C:\Users\dangz\Documents\Fotolog\Aesthetic\descriptors\fish.py", line 97, in <listcomp>
    return np.float32([fisher_vector(image_descriptors(file), *gmm) for file in files])
  File "C:\Users\dangz\Documents\Fotolog\Aesthetic\descriptors\fish.py", line 74, in fisher_vector
    fv = np.concatenate([np.concatenate(a), np.concatenate(b), np.concatenate(c)])
ValueError: zero-dimensional arrays cannot be concatenated

The error seems to be in line 74:
fv = np.concatenate([np.concatenate(a), np.concatenate(b), np.concatenate(c)])
Also, when I check

a = fisher_vector_weights(s0, s1, s2, means, covs, w, T)
b = fisher_vector_means(s0, s1, s2, means, covs, w, T)
c = fisher_vector_sigma(s0, s1, s2, means, covs, w, T)

the means (b) and sigma (c) fisher vectors are all 0s.