Can't use SURF, SIFT in OpenCV

Question

I'm trying a simple thing like

detector = cv2.SIFT()

and get this bad error

detector = cv2.SIFT()
AttributeError: 'module' object has no attribute 'SIFT'

I do not understand that because cv2 is installed.

cv2.__version__ is

$Rev: 4557 $

My system is Ubuntu 12.04.

Maybe someone has got the same problem and could help me.

EDIT:

Long story short, testypypypy.py:

import cv2

detector = cv2.SIFT()

ERROR:

Traceback (most recent call last):
  File "testypypy.py", line 3, in <module>
    detector = cv2.SIFT()
AttributeError: 'module' object has no attribute 'SIFT

If I take SURF it works because SURF is in dir(cv2) but if I also take cv2.BFMatcher() I get the same error... So it's missing and I have to add it but I don't know how.

Yoga

did you look at the output of dir(cv2) to see if SIFT was listed? – Shrapnel

there is only 'SIFT_COMMON_PARAMS_AVERAGE_ANGLE','SIFT_COMMON_PARAMS_FIRST_ANGLE' listed. Can i update "SIFT" and more things like "cv2.BFMatcher()" ??? IF Yes, how ? :) thanks – Yoga

Steps suggested by Marten worked for me. :) I tried it on Ubuntu 12.4 running on AWS. One thing i changed in step 4: use 'sudo make install' instead of 'make install' I was also able to make SIFT & SURF work on Fedora using instructions from link below: g7smy.co.uk/?p=366 – Minstrel

Perhaps you have a newer version of OpenCV: try cv2.xfeatures2d.SURF_create instead of cv2.SURF (many tutorials are using the old call). – Aymara

SIFT and SURF are in the contributed library section and must be included specifically with the install. see docs.opencv.org/4.1.1/d2/dca/group__xfeatures2d__nonfree.html – Eparchy

Answer 1

I think this is far from the "correct" way to do it (the "correct" way on Ubuntu seems to be to stick to a broken and/or outdated OpenCV), but for me building opencv-2.4.6.1 from source brings back cv2.SIFT and cv2.SURF.

Steps:

1. Download opencv-2.4.6.1.tar.gz from opencv.org.

2. Extract the source:

   tar -xf opencv-2.4.6.1.tar.gz -C /tmp
   

3. Configure the source. This will tell OpenCV to install into .opencv-2.4.6.1 in your home directory:

   cmake -D CMAKE_BUILD_TYPE=RELEASE \
         -D BUILD_PYTHON_SUPPORT=ON \
         -D WITH_XINE=ON \
         -D WITH_OPENGL=ON \
         -D WITH_TBB=ON \
         -D BUILD_EXAMPLES=ON \
         -D BUILD_NEW_PYTHON_SUPPORT=ON \
         -D WITH_V4L=ON \
         -D CMAKE_INSTALL_PREFIX=~/.opencv-2.4.6.1 \
         /tmp/opencv-2.4.6.1
   

4. Build and install:

   cd /tmp/opencv-2.4.6.1
   make -j4
   make install
   

5. Set PYTHONPATH (this works in bash, I have no clue about other shells):

   export PYTHONPATH=~/.opencv-2.4.6.1/lib/python2.7/dist-packages
   

Now if I start python and import cv2 (for me, this produces a gnome-keyring warning), I have cv2.SIFT and cv2.SURF available.

Whiting

I've tried this out but got still the same Error. Is it possible to delete opencv fully and will it help?! – Yoga

Is it the new or old cv2 that is loaded? What does print cv2.__version__ say? – Enlightenment

print cv2.__version__ says $Rev: 4557 $ – Yoga

@Linda: Then it loads the old one. Restart python, and try import os,sys, followed by os.chdir(os.path.expanduser('~/.opencv-2.4.6/lib')) and sys.path.append(os.path.expanduser('~/.opencv-2.4.6.1/lib/python2.7/dist-packages')), and then check which version it loads. It should be the new one. – Enlightenment

well its working now!!! Thank you (but its still $Rev: 4557 $ !!! And do i always have to put this lines into my code? Is there a possible way to fix it for ever ? BTW: its os.chdir(os.path.expanduser('~/.opencv-2.4.6.1/lib')) – Yoga

@Linda: All I have to do is to set PYTHONPATH before i start python (i do this in my .bashrc, so it is done on login), and then I can import cv2. I dont have the old and broken opencv though, and it feels like your system is confused as to which to load. If you have sudo rights you could try removing it (you can always reinstall it if you want it back); as far as I understand, the command sudo apt-get remove libopencv-* libcv-* should be enough for this. – Enlightenment

You can do pretty much this same process with openCV 3.4.3 (I did 3.4.3.18), but you need to have the opencv-contrib source as well. Add the following flags to the cmake command: -D OPENCV_ENABLE_NON_FREE=ON and -D OPENCV_EXTRA_MODULES_PATH=/path/to/opencv_contrib-3.4.3/modules. This is what I had when building it on raspbian 4.14.70-v7+ (note I have examples installed too): cmake -D CMAKE_BUILD_TYPE=RELEASE -D CMAKE_INSTALL_PREFIX=/usr/local -D INSTALL_PYTHON_EXAMPLES=ON -D OPENCV_EXTRA_MODULES_PATH=~/opencv_contrib-3.4.3/modules -D BUILD_EXAMPLES=ON -D OPENCV_ENABLE_NONFREE=ON .. – Anglonorman

Answer 2

There is a pip source that makes this very easy.

1. If you have another version of opencv-python installed use this command to remove it to avoid conflicts:

   pip uninstall opencv-python
   

2. Then install the contrib version with this:

   pip install opencv-contrib-python
   

3. SIFT usage:

   import cv2
   sift = cv2.xfeatures2d.SIFT_create()
   

Farland

This worked like a charm. Don't know why this is not upvoted. – Kingdom

When I do that, I get: error: OpenCV(3.4.3) C:\projects\opencv-python\opencv_contrib\modules\xfeatures2d\src\sift.cpp:1207: error: (-213:The function/feature is not implemented) This algorithm is patented and is excluded in this configuration; Set OPENCV_ENABLE_NONFREE CMake option and rebuild the library in function 'cv::xfeatures2d::SIFT::create'. I guess the downloaded version is not compiled with OPENCV_ENABLE_NONFREE – Mani

As of April 2019, the latest version (v4.0.0.21) complains (@Mani pointed out). You can install an older version pip install -U opencv-contrib-python==3.4.0.12 to avoid this. – Selfsame

As of Sep 2019, @Falcon's suggestion worked for me. Basically uninstall opencv-python and install the version 3.4.2.16 of opencv-contrib-phython – Touber

As of February 2020 this does not work. Lowest possible version is 3.4.8.29, and the same error mentioned by @Mani is thrown. – Opisthognathous

I heard that patent for SIFT patent will expire this month or next month, maybe it will come back in OpenCV. – Mani

@Opisthognathous found any solution? – Goulette

@AliİhsanElmas Yep, there's basically no other way than manually building everything. Check my answer : #18562410 – Opisthognathous

Answer 3

For recent information on this issue (as of Sept 2015) consult this page.

Most information on this question here is obsolete.

What pyimagesearch is saying is that SURF/SIFT were moved to opencv_contrib because of patent issues.

For installation there is also a nice page that tells you how to install opencv with opencv_contrib and Python support so you get SURF/SIFT.

Notice that the API also changed. Now it's like this:

sift = cv2.xfeatures2d.SIFT_create()

Before I found the above pages, I also suffered quite a bit. But the pages listed do a very good job of helping with installation and explaining what's wrong.

Kurrajong

cv2.xfeatures2d.SIFT_create() <-- thanks for pointing that out. – Allonym

what about cv2.DescriptorExtractor_create("SIFT")? What will be its substitute? – Geminian

@Allonym How did you get ` cv2.xfeatures2d.SIFT_create()` working? It is still giving me AttributeError: 'module' object has no attribute 'xfeatures2d' . How did you install opencv and opencv_contriv? I did pip install python-opencv and next steps from the page linked in the OP. – Kingdom

This seems a rather lengthy process, on the other hand doing pip install python-opencv and pip install opencv-contrib-python solves the issues immediately. – Kingdom

I too don't see xfeatures2d as available in the current version of OpenCV – Hux

I installed an config referring this learnopencv.com/install-opencv3-on-ubuntu and change to this cv2.xfeatures2d.SIFT_create() do the trick :D – Homelike

Answer 4

FYI, as of 3.0.0 SIFT and friends are in a contrib repo located at https://github.com/Itseez/opencv_contrib and are not included with opencv by default.

Roundtree

Answer 5

for debian users its 'easy' to create their own libopencv-nonfree package.

i followed the opencv tutorial for python, but in my debian the SIFT and SURF modules were missing. And there is no non-free package available for debian including SIFT and SURF etc.

They were stripped from the package due to license issues....

i never created a package for debian before (adding a new module etc) but i followed some small steps in the debian tutorials and tried and guessed around a bit, and after 1 day, voila... i got working a libopencv-nonfree2.4 deb package and a python module with correct bindings.

(i dont know if i also needed to install the newly built python-opencv package or only the nonfree... i re-installed both and got a working python opencv library with all necessary nonfree modules!)

ok, here it is:

!this is for libopencv 2.4!

!you can do all steps except installing as a normal user!

we need the built essesntials and some tools from debian repository to compile and create a new package:

sudo apt-get install build-essential fakeroot devscripts

create a directory in your home and change to that directory:

cd ~ && mkdir opencv-debian
cd opencv-debian

download the needed packages:

apt-get source libopencv-core2.4

and download all needed dependency packages to build the new opencv

apt-get build-dep libopencv-core2.4

this will download the neeeded sources and create a directory called "opencv-2.4.9.1+dfsg"

change to that directory:

cd opencv-2.4.9.1+dfsg

now you can test if the package will built without modifications by typing:

fakeroot debian/rules binary

this will take a long time! this step should finish without errors you now have a lot of .deb packages in your opencv-debian directory

now we make some modifications to the package definition to let debian buld the nonfree modules and package!

change to the opencv-debian directory and download the correct opencv source.. in my case opencv 2.4.9 or so

i got mine from https://github.com/Itseez/opencv/releases

wget https://codeload.github.com/Itseez/opencv/tar.gz/2.4.9

this will download opencv-2.4.9.tar.gz

extract the archive:

tar -xzvf opencv-2.4.9.tar.gz

this will unpack the original source to a directory called opencv-2.4.9

now copy the nonfree modules from original source to the debian source:

cp -rv opencv-2.4.9/modules/nonfree opencv-2.4.9.1+dfsg/modules/

ok, now we have the source of the nonfree modules, but thats not enough for debian... we need to modify 1 file and create a new one

we have to edit the debian control file and add a new section at end of file: (i use mcedit as an editor here)

mcedit opencv-2.4.9.1+dfsg/debian/control

or use any other editor of your choice

and add this section:

Package: libopencv-nonfree2.4
Architecture: any
Depends: ${shlibs:Depends}, ${misc:Depends}
Description: OpenCV Nonfree Modules like SIFT and SURF
 This package contains nonfree modules for the OpenCV (Open Computer Vision)
 library.
 .
 The Open Computer Vision Library is a collection of algorithms and sample
 code for various computer vision problems. The library is compatible with
 IPL (Intel's Image Processing Library) and, if available, can use IPP
 (Intel's Integrated Performance Primitives) for better performance.
 .
 OpenCV provides low level portable data types and operators, and a set
 of high level functionalities for video acquisition, image processing and
 analysis, structural analysis, motion analysis and object tracking, object
 recognition, camera calibration and 3D reconstruction.

now we create a new file called libopencv-nonfree2.4.install

touch opencv-2.4.9.1+dfsg/debian/libopencv-nonfree2.4.install

and edit:

mcedit opencv-2.4.9.1+dfsg/debian/libopencv-nonfree2.4.install

and add the following content:

usr/lib/*/libopencv_nonfree.so.*

ok, thats it, now create the packages again:

cd opencv-2.4.9.1+dfsg

first a clean up:

fakeroot debian/rules clean

and build:

fakeroot debian/rules binary

et voila... after a while you have a fresh built and a new package libopencv-nonfree2.4.deb!

now install as root:

dpkg -i libopencv-nonfree2.4.deb
dpkg -i python-opencv.deb

and test!

import cv2
import numpy as np
from matplotlib import pyplot as plt

img = cv2.imread('test.jpg')
gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)

sift = cv2.SIFT()
kp = sift.detect(gray,None)
img=cv2.drawKeypoints(gray,kp)

corners = cv2.goodFeaturesToTrack(gray,16,0.05,10)
corners = np.int0(corners)

for i in corners:
    x,y = i.ravel()
    cv2.circle(img,(x,y),90,255,3)

plt.imshow(img),plt.show()

have fun!

Tranquilizer

This worked for me as well on Ubuntu 14.04. I can now use cv2.sift in Python. – Fingered

Answer 6

IN opencv3.x SIFT() & SURF() are no longer exist .for this

uninstall all the opencv versions

python -m pip uninstall opencv-python

python -m pip uninstall opencv-contrib-python

after that install opencv-contrib to include sift() and surf() using below given command with python(3.x)

python -m pip install opencv-contrib-python==3.4.2.16

then you can use

sift = cv2.xfeatures2d.SIFT_create()

Darrickdarrill

Thank you sir! This is a huge contribution – Replica

This. This is all it takes at least on Windows. – Pershing

It worked for me just needed to modify it for python3 in RPI 4. But not working in windows. In windows it's saying that version not found: "python3 -m pip install opencv-contrib-python==3.4.2.16" – Cancer

Answer 7

As an Anaconda user, I wanted to find one or two appropriate commands to solve the problem. Fortunately, this answer helped. For conda 4.5.11 (use conda -V to check Anaconda version) I have performed next steps:

# Python version does not matter, most likely, check yourself
conda create -n myenv python=3.6     
conda activate myenv
conda install -c menpo opencv

That will install OpenCV 2.4.11. Anaconda's another command conda install -c menpo opencv3 will install OpenCV3, but Python has to be downgraded to 2.7. To install OpenCV3 with Python3 use next (due to the first link):

conda create -n myenv python  
pip install opencv-python==3.4.2.16
pip install opencv-contrib-python==3.4.2.16

Check the SIFT:

conda activate myenv
python
>>> cv2.xfeatures2d.SIFT_create()
<xfeatures2d_SIFT 000002A3478655B0>

Shiest

I have just settled on the Ubuntu and checked installation steps above with pip3 instead of pip. All works great for version 3.4.2.16. Version 4.0.0.21 restricts using SIFT and requires compilation from the source for using SIFT. – Shiest

Answer 8

This is what worked for me in September 2020:

1. Remove previous versions:

   pip uninstall opencv-python
   pip uninstall opencv-contrib-python
   

2. Install this particular version (3.4.2.16 was no longer available)

   pip install opencv-python==3.4.2.17
   pip install opencv-contrib-python==3.4.2.17
   

3. Create the detectors with:

   import cv2
   sift = cv2.xfeatures2d.SIFT_create()
   surf = cv2.xfeatures2d.SURF_create()
   

Linton

Answer 9

just change SHIFT to ORB, I think it make occur because of non-relevant version, ORB is efficient and better alternative of SHIFT or SURF.

As I also face same problem when i was used cv2.SHIFT()

ERROR: AttributeError: 'module' object has no attribute 'SIFT'

Now its completely working for me please try this:

ORB = cv2.ORB()

Braunschweig

Answer 10

since I had already compiled opencv when I discovered this problem, all I had to do was (from my opencv build directory):

make opencv_nonfree
sudo make install

Ooze

Answer 11

Install Python opencv

pip install opencv-python

and instead of using ..

cv2.SIFT()

Use

cv2.SIFT_create()

working code using opencv-python below

import cv2
img1 = cv2.imread('yourimg.png',0)
sift = cv2.SIFT_create()
kp1, des1 = sift.detectAndCompute(img1,None) #keypoint and descriptors
...

you can also install "opencv-contrib-python" and use "cv2.xfeatures2d.SIFT_create()" but that is secondary and up to you.. working code using the python package opencv-contrib-python

import cv2
img1 = cv2.imread('yourimg.png',0)
sift = cv2.xfeatures2d.SIFT_create()
kp1, des1 = sift.detectAndCompute(img1,None) #keypoint and descriptors

Thanks

Cudbear

Answer 12

Follow this installation operation

cmake -D CMAKE_BUILD_TYPE=RELEASE -D CMAKE_INSTALL_PREFIX=/usr/local
    -D WITH_TBB=ON -D BUILD_NEW_PYTHON_SUPPORT=ON -D WITH_V4L=ON 
    -D INSTALL_C_EXAMPLES=ON -D INSTALL_PYTHON_EXAMPLES=ON 
    -D BUILD_EXAMPLES=ON -D WITH_QT=ON -D WITH_OPENGL=ON ..

using this command will install library to your /usr/local/lib.

Phlegmatic

Please note that not everyone has write access outside their home directory (this is especially true on campus computers). Installing at /usr/local/lib typically would require sudo rights. – Enlightenment

Answer 13

1. Install OpenCV-Contrib

2. import cv2 sift = cv2.xfeatures2d.SIFT_create()

3. sift.something()

This is the easy way to install the Contrib.

Elroyels

It would be nice to provide some explanation to accompany your code, particularly point 2. Also note your formatting is a little off on number 3. – Diphenylhydantoin

Answer 14

Just change the version of opencv to 3.4.2.16 .Since it is patented it is not available in newer version.

Safier

Answer 15

You can proceed in this way. It must work for you as well!

Step 1:

virtualenv venv 
source venv/bin/activate

---

Step 2:

sudo python3 -m pip install opencv-python==3.4.2.16 
sudo python3 -m pip install opencv-contrib-python==3.4.2.16

---

Step 3:

import cv2
sift = cv2.xfeatures2d.SIFT_create()

---

Don't use cv2.SIFT() . It will raise an exception.

Hotien

Answer 16

It is now back to normal opencv, thanks to the expiration of the patent! Check this: https://docs.opencv.org/4.5.4/da/df5/tutorial_py_sift_intro.html As long as your opencv version is 4.4.0+!

Summit

Answer 17

The approach suggested by vizzy also works with OpenCV 2.4.8, as when building the non-free package under Ubuntu 14.04 LTS.

This dependency issue may prevent installation of the non-free package:

 libopencv-nonfree2.4 depends on libopencv-ocl2.4; however:
  Package libopencv-ocl2.4 is not installed.

Easily fixable because the missing package can be installed from the ones just built:

dpkg -i libopencv-ocl2.4_2.4.8+dfsg1-2ubuntu1_amd64.deb

After that the install proceeds as explained in vizzy's answer.

Vaginismus

Answer 18

None of the above suggested solutions worked for me. I use Anaconda and found that opencv version 3.3.1 still had Sift enabled. If you want to test in isolated conda environment, try the following inspired from @A.Ametov's answer above

conda create -n testenv opencv=3.3.1     
conda activate testenv

conda activate myenv
python
#Check version of opencv being used
>>> import cv2
>>> cv2.__version__
#Check if Sift is available
>>> cv2.xfeatures2d.SIFT_create()
<xfeatures2d_SIFT 000002A3478655B0>

Rubato

Answer 19

Change this:

sift = cv2.xfeatures2d.SIFT_create()

By this:

cv2.ORB_create()

Anteroom

Answer 20

The following page provides a relatively good guide that requires few corrections : https://cv-tricks.com/how-to/installation-of-opencv-4-1-0-in-windows-10-from-source/

On step 8, when you are choosing generator for the project (tool used for building), don't forget to specify x64 in the second field if you need it. If you don't, you will get LNK1112 error, which is a linker error caused if one module is created with the x64 compiler, and another module is created with the x86 compiler.

Next, when choosing flags in step 9, don't forget to tick the "OPENCV_ENABLE_NONFREE", and set the path for flag "OPENCV_EXTRA_MODULES_PATH" flag. The path must be set to "modules" folder in opencv-contrib-python.

So you need both opencv, and opencv-contrib-python in order to use SIFT and SURF.

Opisthognathous

Answer 21

As per June 2020, I suppose this version (pip install -U opencv-contrib-python==3.4.2.16) is still working. so install it and enjoy.

Yoder

Answer 22

try this

!pip install opencv-contrib-python==4.4.0.44 sift = cv2.SIFT_create()

Rancid