ximage-io claims to use the shape convention WIDTH x HEIGHT x CHANNELS (see http://xtensor-io.readthedocs.io/en/latest/api_reference.html#images) along with the usual c-order strides (increasing from back to front). However, this is wrong.
All file formats store images such that the WIDTH axis has smaller stride than the HEIGHT axis, and xtensor-io reads the memory unchanged. This means that the shape must be HEIGHT x WIDTH x CHANNELS with corresponding strides. All other libraries that follow the c-order convention use this shape, see for example scikit-image. You can easily verify the error with a simple program:
auto in = xt::load_image("color_image.tif");
for(int y=0; y<in.shape()[1]; ++y)
{
in(100, y, 1) = 255;
}
xt::dump_image("result.tif", in);This should draw a vertical line, but actually draws a horizontal one.
ximage-io claims to use the shape convention
WIDTH x HEIGHT x CHANNELS(see http://xtensor-io.readthedocs.io/en/latest/api_reference.html#images) along with the usual c-order strides (increasing from back to front). However, this is wrong.All file formats store images such that the WIDTH axis has smaller stride than the HEIGHT axis, and xtensor-io reads the memory unchanged. This means that the shape must be
HEIGHT x WIDTH x CHANNELSwith corresponding strides. All other libraries that follow the c-order convention use this shape, see for example scikit-image. You can easily verify the error with a simple program:This should draw a vertical line, but actually draws a horizontal one.