Running examples

[pha-nguyen]

Hi,

Thanks for your great work!

I can see the implementation of your core GridShift algorithm, and I am diving into the applications that you mentioned in the paper, i.e. object tracking, and segmentation. Could you please provide examples of how to apply your algorithm to those tasks?

Appreciate!

[Treeboy2762]

Hi, I hope you have already solved the problem.

The below code is a reproducible example.

from GridShiftPP2 import GridShiftPP
import numpy as np

gs = GridShiftPP(0.25, 0.001, 10)

data = np.array([[0,0], [1,1]])
result = gs.fit_predict(data)

print(result)

[javiabellan]

I can not make this color quantization code to work:

import GridShiftPP2  # This is the name given in the setup.py file
import cv2
import requests
import numpy as np
from matplotlib.pyplot import imshow

img_url = "https://t2.gstatic.com/licensed-image?q=tbn:ANd9GcRS7Outpkdd8Ir9TmzQnF5HxJr6nIhJIl2Cgp0mkLtlzF4zSRCx5Wa2bbKkgkUbp741Rso7ZYl90gzJmke9bkE"

response = requests.get(img_url)
img_np_array = np.frombuffer(response.content, np.uint8)
img_bgr = cv2.imdecode(img_np_array, cv2.IMREAD_COLOR)
img_rgb = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2RGB)

def color_qunatization(img, bandwidth=1.0, threshold=0.001, iterations=10):

    height, width, c = img.shape
    img_feats = img.reshape(-1,3) / 255

    # GridShift
    grid_shift_instance = GridShiftPP2.GridShiftPP(bandwidth=bandwidth, threshold=threshold, iterations=iterations)
    label_map, cluster_centers = grid_shift_instance.fit_predict(img_feats)

    label_map = label_map.reshape(height, width)

    colors = cluster_centers # float64 of shape [N_clusters,3]

    img_q_rgb = colors[label_map] * 255
    img_q_rgb = img_q_rgb.round().astype(np.uint8)

    return label_map, colors, img_q_rgb

label_map, colors, img_q_rgb = color_qunatization(img_rgb)
imshow(img_q_rgb)

All i got is this single avg color:

Note that Lab color space is recommended for color distances, but i keep the RGB for the simplicity of the code.