Closed ashkart closed 8 years ago
Hi @ashkart , Can you provide the detailed build message, your platform info, your usage message, and specific errors? To use libvot_feature, you should have opencv 2.4 installed. I have also updated the master branch a bit. You can try again. Thanks.
Hi @hlzz , My platform is Linux mint 17.3. Opencv 2.4.13. I am using those instructions u put on main project's page to compile the libvot. There is no any errors on compiling. Tests are fine too. `Test project /home/max/SDKs/libvot-master/build Start 1: unit_test 1/3 Test #1: unit_test ........................ Passed 0.01 sec Start 2: opencv_test 2/3 Test #2: opencv_test ...................... Passed 0.19 sec Start 3: libvot_feature_test 3/3 Test #3: libvot_feature_test .............. Passed 0.95 sec
100% tests passed, 0 tests failed out of 3
Total Test time (real) = 1.16 sec `
When I try to use image_feature, I do this
./libvot_feature car.jpg
The image called "car.jpg" is in same directory with libvot_feature executable (build/bin/qwe) wich i created.
Console input in attach.
log.txt
And what folder contains now...
I tryed to research the problem by myself, but i cant compile project in my IDE (netbeans and codeblocks), i get some errors in output, wich solutions i even cant find in google 0_0.
hi @ashkart , libvot_feature takes a image filename list as input. For a single image case, you can generate a image list file containing the path to car.jpg. The image list has only one line:
/home/max/SDKs/libvot-master/build/bin/qwe/car.jpg
And pass that image list to libvot_feature, like ./libvot_feature image_list.txt
Oooohh.... now it runs fine. And I successfully created a tree. Dont understand how to use it yet, but it seems all stuff works fine. Thanks a lot!
@ashkart , You're welcome. I will add a section to guard against single image input. Thanks.
I've created some database with image_search executable example on those sift bins I got with libvot_feature executable. How to send query image to that database? I know it is offtop, but there is no wiki about it, and no any documentation((
Hi @ashkart , If you mean update the image database, it is not well supported now but the api is flexible and we should be supporting it in the future. If you mean query an image, you can check out web_search in example.
I mean send query in DB (query containing an image) and receive an answer - is given image (or matching image) exists in DB.
@ashkart You can first extract the sift features from query images and database images, then run image_search. It consists a complete pipeline containing training, image DB building and querying. The results are store in the output_folder you specified.:)
Okay. I extracted sift data from images i need to store in DB. Executed the image_search, and got results in folder: Now I will use one of existing sift file in my query. Should i make a list with that sift and run image_search again? It clears up files which are on screen below and puts there info about this one sift...
No. If you have included the query sift file in the image database building process. You are good to go. The query results are in match.out
If you want to do a simple single-image query, check out web_search :)
What that web_search query returns in success case?
For me it looks like that output in attach. attach.txt
@ashkart It will output a rank list file (*.rank) in the output_folder, specifying the top-k retrieval result, from top to bottom
is my input correct?
./web_search single vocab_out/match_pairs.txt vocab_out/db.out ouput
single - is a txt with path to fist of query image. vocab_out/match_pairs.txt - list of images paired with sift in db folder vocab_out/db.out - db file ouput - output folder for web_search
Nope. I will give you an example: (type this in libvot/build/bin) ./web_search ../../test_data/0001.desc ../../test_data/list vocab_out/db.out
This command will generate a rank list called 0001.rank in the same folder as the executable file, which contains the retrieval results. For the usage of additional flags, you can type the following to see: ./web_search --helpshort
oooohh.... and what is the .desc files?)
It the feature descriptors generated by openmvg, an alternative way to use libvot when libvot was lack of a feature detector. You can find them in test_data. You can also replace .desc files with the .sift files generated by libvot_feature.
and what is non-alternative way in my case? the primary one )) i didnt use open mvg, that means i dont need those .desc files. i cant use them now. there is no my images data there.
@ashkart
You can also replace .desc files with the .sift files generated by libvot_feature.
oh, sorry, thanks, will try it now))
yes, it works! man, dont stop working on ths project. quick image recognition search is very expensive pleasure. in my country there is no such projects.
It would be nice to see matching scores in output
@ashkart Thanks. We are working towards the next release. Feel free to fork it and play with it. :)
image_feauture example doesnt work. Files it creates are named with unreadable encoding. And i cant try image_search because of that.