Coords opts

[d-v-b]

This addition adds the ability to get the set of (x,y) points inside a polygon drawn in an instance of lightning.image

To test this new functionality, draw a region on an image and call the following:

# get set of pixels from image
coords = viz.get_coords(as_points=True)[0]
# build an array with the same size in (x,y) as the image data
tmp = np.zeros(dims[0:2])

# for each pixel, color the image at that point.
for pix in coords:
        tmp[pix[0],pix[1]] = 1

# the resulting binary image should match the shape and position of the drawn polygon
plt.imshow(tmp)

[freeman-lab]

tolist() is not in place, make sure to call it on the object that will be returned.

[freeman-lab]

@d-v-b this looks great! I think it would be nice to add one more output type, a mask for each region, of 0s and 1s with 1s where the region was defined. As we'll then have three options, I would suggest the following change to the name / function definition:

def get_regions(self, return_type='bounds', dims=None):

Where the three options are bounds, points, and mask. Then the core of the method would look like:

    coords = user_data['coords']
    if return_type == 'bounds':
        return coords
    if return_type == 'points':
        return [get_points_from_polygon(x) for x in coords]
    else if return_type == 'mask':
        if not dims:
            raise Exception('Must provide dimensions to return masks')
        # call function to generate a mask on each region
    else:
        raise Exception('Option %s not supported' % return_type)

Want to give this a try?

[d-v-b]

@freeman-lab yep, I can work those suggestions in. Expect something along those lines in a few hours

[d-v-b]

@freeman-lab:

A list of boundary points for each polygon are returned when get_coords is called like this:
bounds = viz.get_coords()

Given dims, the dimensions of the image containing the polygons, a list of binary masks formed by each polygon is returned when get_coords is called like this:
mask = viz.get_coords(return_type='mask',dims=dims)

A list of lists of points internal to each polygon is returned when get_coords is called like this:
pts = viz.get_coords(return_type='points',dims=dims)

The argument dims is required for both inner points and masks because the masks are generated by calling the function bounds_to_mask(bounds,mask) on the bounds, and the points are subsequently generated by calling mask_to_points(mask)

[freeman-lab]

Remove line break

[freeman-lab]

@d-v-b looking very close! Made a couple formatting suggestions, only other thing is that I'd rather use the get_points_from_polygon from before instead of mask_to_points(polygon_to_mask()), becuase the earlier one didn't require knowing the dims. Normally I'd prefer the modularity and consistency of your approach, but I think the ability to get points without providing the dimensions is actually nicer to have, and you need an extra function in either case.

[d-v-b]

@freeman-lab now get_coords(return_type='points') does not require dims as an argument, thanks to the addition of a function polygon_to_points

for consistency in naming, I also re-named the mask-generating function to polygon_to_mask.

[freeman-lab]

@d-v-b Terrific, LGTM, really nice work! Merging it in.