A difference between Halide and Liszt is that Halide doesn't make decisions for the programmer, but Liszt does.
vasua
I'm curious why Halide isn't more popular, if it offers so many performance benefits. In most, if not all, all of the CV applications I've seen, people have used OpenCV in C++. It does have a GPU backend in addition to the CPU one, but I'm almost positive that OpenCV doesn't perform any of these sorts of applications. If Halide is as easy to use as it seems, and offers the massive performance benefits, why has it not taken over the research and commercial space for computer vision tasks?
I will certainly be trying Halide for my next CV projects, though :)
Firephinx
@vasua I think many people choose OpenCV over Halide because not many people in the world know the underlying algorithms that are already implemented OpenCV. OpenCV is great for quick prototyping by just linking up the inputs and outputs of different functions and knowing at a high level what the functions do. Halide is more for those who are actually researching in Computer Vision, which is why it is popular in Google and other Computer Vision labs including the Graphics and Computer Vision labs at CMU. I'm sure that as more and more people become educated in Computer Vision like CMU's new Masters in Computer Vision program, Halide will be increasingly used; however, there are simply not enough people in the world that have the expertise to use it right now.
I agree that I'd like to use Halide more, but that means I would also have to read many different computer vision papers, and reimplement their algorithms from scratch.
A difference between Halide and Liszt is that Halide doesn't make decisions for the programmer, but Liszt does.
I'm curious why Halide isn't more popular, if it offers so many performance benefits. In most, if not all, all of the CV applications I've seen, people have used OpenCV in C++. It does have a GPU backend in addition to the CPU one, but I'm almost positive that OpenCV doesn't perform any of these sorts of applications. If Halide is as easy to use as it seems, and offers the massive performance benefits, why has it not taken over the research and commercial space for computer vision tasks?
I will certainly be trying Halide for my next CV projects, though :)
@vasua I think many people choose OpenCV over Halide because not many people in the world know the underlying algorithms that are already implemented OpenCV. OpenCV is great for quick prototyping by just linking up the inputs and outputs of different functions and knowing at a high level what the functions do. Halide is more for those who are actually researching in Computer Vision, which is why it is popular in Google and other Computer Vision labs including the Graphics and Computer Vision labs at CMU. I'm sure that as more and more people become educated in Computer Vision like CMU's new Masters in Computer Vision program, Halide will be increasingly used; however, there are simply not enough people in the world that have the expertise to use it right now.
I agree that I'd like to use Halide more, but that means I would also have to read many different computer vision papers, and reimplement their algorithms from scratch.