test.py

from quickmatch_cluster import *
import cv2


def get_size(path):
    size = {}
    with open(path, "r") as f:
        lines = f.readlines()
        for line in lines:
            line = line.split(' ')
            size[line[0]] = [int(line[1]), int(line[2])]
    return size


def get_ground_truth(m, n):
    """

    :param m: 6
    :param n: 13
    :return:
    """
    ground_truth = []
    for i in range(n):
        ground_truth.append([j*n+i for j in range(m)])
    return ground_truth


def eval(pred, ground_truth):
    num = len(ground_truth)
    correct = 0
    for p in pred:
        if sorted(p) in ground_truth:
            correct += 1
    return correct / num


def visualize():
    for i in range(6):
        cv2.namedWindow("imgs[%d]" % (i + 1), cv2.WINDOW_NORMAL)
    imgs = np.zeros((6, col, row, 3), dtype=np.uint8)
    for c in clusters:
        for i in range(6):
            img_path = data_path + "\\%s\\img%d.ppm" % (categroy, i + 1)
            imgs[i] = cv2.imread(img_path, cv2.IMREAD_COLOR)

        for idx in c:
            i = idx // n
            x, y = kp[idx][0], kp[idx][1]
            cv2.circle(imgs[i], (x, y), 8, (0, 255, 255), 6)

        for i in range(6):
            cv2.imshow("imgs[%d]" % (i + 1), imgs[i])
        cv2.waitKey(0)


print(get_ground_truth(6, 13))
data_path = "C:\\Users\\ChenZixuan\\OneDrive\\Document\\Research\\Graph_Matching\\Graffiti_dataset"
categroy = 'graf'
size = get_size(data_path + "\\size")
row, col = size[categroy]

f = np.load(data_path + '\\%s\\%s.npz'%(categroy, categroy))
kp, des = f['kp'], f['des']
n, m, t = des.shape
# n is number of points per image
# m is number of image
data = des.transpose(1, 0, 2).reshape(m*n, t)
n, m, t = kp.shape
kp = kp.transpose(1, 0, 2).reshape(m*n, t)

clusters = quickmatch_cluster(data, n, kernel=np.exp, flag_low_memory=False)

ground_truth = get_ground_truth(m, n)
correct_rate = eval(clusters, ground_truth)
print("ground_truth_num=%d, accuracy:%.4f" % (n, correct_rate))
visualize()