Re: [julia-users] Computer Vision Package
On Monday, August 10, 2015 01:49:08 PM Mehmet Cem Güntürkün wrote: > However, for that I will need Scharr Operator. Do we have that operator in > Images.jl? See http://timholy.github.io/Images.jl/function_reference.html and search for "sobel" (which is close). But if you want a different one, sure, that would be great to add. --Tim
Re: [julia-users] Computer Vision Package
> but I might still need a layer between Images.jl and my package because > most of tracking algorithm uses only grayscale images, but still not sure > about that. > Based on my experience (re)implementing active appearance models, I'd say many tracking algorithms work even better with colored images than with grayscale ones. Anyway, you can always convert data to a format of your choice (e.g. 2D or 3D arrays), process in the algorithm and convert the result back to Image.
Re: [julia-users] Computer Vision Package
Hey everyone, thanks a lot for your answers and suggestions. That seems I have too many things to learn. *@Tim* Actually using Images.jl for input, output and core data structures is a good idea because I guess it is already handling many different image formats/extensions but I might still need a layer between Images.jl and my package because most of tracking algorithm uses only grayscale images, but still not sure about that. I guess I should start working on that ASAP to see how it's working. Also for filtering stuff, I am going to implement first: http://robots.stanford.edu/cs223b04/algo_tracking.pdf that is also what OpenCV has for calcOpticalFlowPyrLK, I guess that would be good start point. However, for that I will need Scharr Operator. Do we have that operator in Images.jl? Otherwise I can implement that operator on Images.jl. because If I am going to use Images.jl for filtering and pyramids, I guess all filtering logic should be there. Correct me if I'm wrong. *@Kevin *I didn't see VideoIO.jl, but it seems pretty cool. I will check that but as far as I can see that will solve video input problem. Thanks again. Mehmet -
Re: [julia-users] Computer Vision Package
On Sunday, August 09, 2015 05:26:33 PM Kevin Squire wrote: > I'll second everything that Tim says, except that, for reading video, > VideoIO.jl would probably be more appropriate. Definitely! It's an amazing package. --Tim
Re: [julia-users] Computer Vision Package
I'll second everything that Tim says, except that, for reading video, VideoIO.jl would probably be more appropriate. My general workflow has been to read in images using VideoIO, and then manipulate them using Images.jl. I'll add that Tim has spent a lot of time considering how to make Images.jl very fast (choosing fast algorithms, optimizing for cache hits, etc., and including such optimizations in Julia itself when it makes sense), so I would strongly suggest using his code when feasible. > >- Tracking Algorithms > >- Feature Extraction and Detection > > These would be awesome additions. I like the idea of a separate package. > I'm definitely interested in these areas specifically, so if you start packages for either, I'd be quite interested in contributing. Kevin
Re: [julia-users] Computer Vision Package
Better support for ComputerVision would be awesome. And having separate packages for specialized functionality is a good idea. On Sunday, August 09, 2015 01:48:49 PM Mehmet Cem Güntürkün wrote: >- Create a core package for data structures, input and output. I don't understand why you wouldn't use Images.jl for this. It's quite mature and well-tested, supports a huge array of formats and transformations among them. You can use Images.jl to handle the core image I/O, basic representations and changes in representation. In open source, it's best to make the packages interoperate, because then you leverage the contributions of different people to create something greater than the parts (see https://github.com/JuliaLang/Color.jl/issues/101 for a problem that we're just trying to face now because of this issue). >- Some image processing stuff such as Filters, Pyramids, etc. Images has some quite fast algorithms for gaussian filtering, imfilter, and imfilter_fft. For pyramids, its `restrict` is also really fast. >- Tracking Algorithms >- Feature Extraction and Detection These would be awesome additions. I like the idea of a separate package. Best, --Tim
[julia-users] Computer Vision Package
Hey everyone, I have just completed my first year of Master Degree in Computer Science, Computer Vision to be specific and currently I am using C/C++ for development. However, I really want to use Julia, it is as fast as C and as far as I see, it is much more easier to parallelize operations. The reason why I am not going to use for my Master Thesis Project is that it seems that there is no stable library I can use such as OpenCV. I am not really fan of OpenCV because it is too hard to debug the code. I have read and searched over the blogs and google groups and seen some topics about Computer Vision library but - one of them was wrapping OpenCV library. - some of them decided to use Images.jl which satisfies their requirements. - some of the topics are kind of expired. This is the reason why I want to create new topic about that issue to take some attentions whether anyone has already progressed work that I can join and contribute. Otherwise, I am going to build a package for Computer Vision in pure Julia. Basically my initial plan is: - Create a core package for data structures, input and output. - Some image processing stuff such as Filters, Pyramids, etc. - Tracking Algorithms - Feature Extraction and Detection In addition to these, I want to add Machine Learning procedures such as Decision Trees, Boosting, Neural Networks, etc. Please correct me if I wrote something wrong and I really appreciate any advices and suggestions about this issue because it will be my first open source project. and I want something that is used and make some people's lives easier. Btw, I will call it Owl.jl. Thanks a lot. Mehmet -
[julia-users] Computer vision package?
Anyone know of or is working on a computer vision package? eg, something that can compute common image features like hogs, perform various normalizations, etc.
