I've been implementing my own HoG transform looking at different sources. While the implementation in scikit-image seems to lack certain features (multiple normalization schemes, general block overlap, Gaussian block window, trillinear interpolation/weighting of bin assignments,...) these don't seem to be that important, at least when applied to my current problem (eye blink analysis). Most of these would increase complexity, giving the implementation a complicated look, with little gain. I've also been looking into some practical improvements - integral histogram, separating the cell-block histogram feature to use it with other dense feature transforms such as LBP, a HoG visualization function that would render the visualization at higher resolutions that the original image. Would any of these be welcomed additions to scikit-image? Regards, Martin Savc PuppySaturation
Dear Martin On Tue, Jan 27, 2015 at 11:18 AM, Martin Savc <martin.savc@gmail.com> wrote:
Most of these would increase complexity, giving the implementation a complicated look, with little gain. I've also been looking into some practical improvements - integral histogram, separating the cell-block histogram feature to use it with other dense feature transforms such as LBP, a HoG visualization function that would render the visualization at higher resolutions that the original image.
Personally, I have never used our HoG implementation, but I am very glad that you are doing a thorough review of it! If you can make any improvements (such as the fixes you've already submitted), those are most welcome. Regards Stéfan
Hello, Martin I find out that skimage HOG doesn't have "overlapping feature". May I know whether your patches has merged into master tree or could you please share your branch? Best Regards Dai Yan 在 2015年1月28日星期三 UTC+8上午3:18:34，Martin Savc写道：
I've been implementing my own HoG transform looking at different sources. While the implementation in scikit-image seems to lack certain features (multiple normalization schemes, general block overlap, Gaussian block window, trillinear interpolation/weighting of bin assignments,...) these don't seem to be that important, at least when applied to my current problem (eye blink analysis).
Most of these would increase complexity, giving the implementation a complicated look, with little gain. I've also been looking into some practical improvements - integral histogram, separating the cell-block histogram feature to use it with other dense feature transforms such as LBP, a HoG visualization function that would render the visualization at higher resolutions that the original image.
Would any of these be welcomed additions to scikit-image?
Regards, Martin Savc PuppySaturation
Hello, Martin I am seeking for HOG implementation with "overlap feature". May I know whether you have commited your patches or could you please share your branch? Thanks Best Regards Dai Yan 在 2015年1月28日星期三 UTC+8上午3:18:34，Martin Savc写道：
I've been implementing my own HoG transform looking at different sources. While the implementation in scikit-image seems to lack certain features (multiple normalization schemes, general block overlap, Gaussian block window, trillinear interpolation/weighting of bin assignments,...) these don't seem to be that important, at least when applied to my current problem (eye blink analysis).
Most of these would increase complexity, giving the implementation a complicated look, with little gain. I've also been looking into some practical improvements - integral histogram, separating the cell-block histogram feature to use it with other dense feature transforms such as LBP, a HoG visualization function that would render the visualization at higher resolutions that the original image.
Would any of these be welcomed additions to scikit-image?
Regards, Martin Savc PuppySaturation
participants (3)
-
Dai Yan -
Martin Savc -
Stefan van der Walt