findContours withHierarchy=true segmentation faults on a zeros image

[electronicwhisper]

When I run findContours withHierarchy=true, it works most of the time but segmentation faults if the image I pass in is all zeros.

Below is a test case that shows this behavior. Note that if you make one of the image pixels white it works.

local cv = require "cv"
require "cv.imgcodecs"
require "cv.imgproc"


local img = torch.ByteTensor(400, 400)

-- Uncomment below and this works.
-- img[20][20] = 255


local contours, hierarchy = cv.findContours{
  image=img,
  withHierarchy=true,
  mode=cv.RETR_CCOMP,
  method=cv.CHAIN_APPROX_SIMPLE
}

print(contours)
print(hierarchy)

[shrubb]

Hi, thanks for reporting! This sample works in C++, so there must be a bug in the bindings. Investigating, stay tuned...

[shrubb]

Indeed, there was a (yet another) bug in Tensor conversion. Fixed, thank you.