On 12/24/2013 03:57 AM, Dried Meat wrote:
> Thanks for the reply Samual!
Thanks to you!
> First thank you for this project, I'm surprised you haven't had others
> helping out.
Well, I think it has a lot to do with the lack of "marketing".. It'll
probably get better once I get a domain name and switch to GitHub. And
there's also the fact that there has been so many attempts at coming up
with something useful in the past that most people have simply given up,
or gone back to do raw JNI, and won't be convinced until shown something
that actually works.
> I am not too sure what I will fully need, but what does the Java portion
> of OpenCV have that JavaCV does not?
The APIs are obviously not the same, many little differences there, but
I think the only big thing is a hack to access the camera from native
code on most Android devices. But that doesn't work on all devices,
while Android already provides a standard Java API to capture from
cameras, so I'm not sure that it's actually useful.
> It also seems that you're saying if I do NOT need things like FFmpeg or
> the others it's best to use OpenCV?
Probably, but opencv-java doesn't support callbacks or direct NIO
buffers, for example. Java is a second class citizen in OpenCV. Or
actually third class, Python is second class, and we can't even do
everything with Python, because it's just so slow. Only C++ is fully
supported. Its Java API might never support GPU calls, for example,
because it can't be used on Android anyway, or direct NIO buffers,
because they are slow on Android anyway. "And who programs in Java
outside of Android?" is what most people erroneously believe in the
computer vision community, in my experience, unfortunately.
> Like you mentioned I REALLY am NOT looking to convert C++ code to Java.
>
> The examples are in Scala for JavaCV though, but that is closer to
> Java(and can be used with Java) than C++.
>
> As for the Parser it seems your main purpose is to convert the c++ into
> Java so we can use All of the features in OpenCV?
No, it's purpose is to convert C++ header files (almost)
/automatically/. It's already possible to do it /manually/, and that's
what is in JavaCV at the moment.
> I guess I will ask here that I really am looking to use image and video
> recognition, as well as image stitching with both video and images. I
> am looking to recreate scenes and be able to identify the objects within
> that scene.
>
> Hopefully JavaCV has what I'm looking for, and I would appreciate if you
> could let me know if it is possible to do what I am looking for..
Sure, it sounds like it's what you are looking for :) You are probably
going to need to train it to recognize your objects though:
http://docs.opencv.org/doc/user_guide/ug_traincascade.html
> It does seem better to use an officially supported language, but I
> figure if this is supported for awhile it would be a great choice, I
> definitely do not want to have to convert code later on if this project
> somehow dies.
Well the API might change a bit with the "Presets"... What I really want
for now is to show that it's technically feasible (no reason to believe
that it isn't at this point), and then after that it should be smooth
sailing
> Thank you, all the best, and I hope you(if you want) get some help in
> making this library the best!
Of course! What do you have in mind?
Samuel