selective blur tutorial

Import image

The first step consists in selecting the image, by assigning its name and extension to the variable image_name.

By default, the images will be searched in the images folder.

image_name = "model.jpg"
image_path = os.path.join("images/inputs", image_name)

Mask

Selection

s = Segmenter(image_path)
s.select_from_image()

Select all the subjects that you want to segment by clicking on them (a blue cross will appear on them). Then click on “Submit”.

Then, three candidate masks will appear (mask 0, 1, and 2), as follows:

You can choose the mask that you prefer by writing the number of your choice inside s.choose_mask().

In this case, we will select mask 0:

s.choose_mask(0)

Now we can plot the original image and its masked version:

show_image_and_mask(s.image_bgr, s.best_mask)

Cleaning

In this case the mask is already pretty good, so not much cleaning is needed. However, when necessary, two tools can be used.

The first one is auto_denoise, which removes the areas masked by the model but that are not really related to the subject that we want to isolate. The denoised mask with the selected level is saved automatically to the editor object.

editor = MaskEditor(s.best_mask)
editor.auto_denoise()

Manual edit is the second tool, which allows to manually select areas by drawing rectangles on the image that will be added or removed from the mask. In this case, we are adding to the mask a small black point that has not been selected by the model. Add-Mode is active when its color is dark-grey; in this case, if we click on it, we switch to remove mode.

The edited mask is saved only when Save is clicked. “Submit” can be used to see a preview of the changes, which will be saved only when “Save” is clicked.

editor.manual_edit()

Blur

Generate depth map

When the image is loaded into the Selector class, the MiDaS model produces a depth map of it that can be viewed with show_depth_map().

sel = Selector(image_path, model="DPT_Hybrid", mask=editor.mask)
sel.show_depth_map()

Apply blur

The function select_kernel_size allows to select interactively the desired level of blur (for more details, see the function documentation).

Final image

The final step consists in blending the blurred image with the masked subject. When the maximum sharpness level is chosen, as in this case, the masked subject will not be blurred.

b = Blender(mask=editor.mask, original_image=sel.image, image_blurred=sel.image_blurred)
b.blend()