create_similarityMatrix_cls_object_ade.py

import numpy as np
from scipy import misc

def softmax(x):
    ex = np.exp(x)
    sum_ex = np.sum(np.exp(x))
    return ex/sum_ex

ADE_gt_tr = './ade/ADEChallengeData2016/ADE_gt_tr.txt'
img_path_list_tr = []
img_label_list_tr = []
with open(ADE_gt_tr, 'r') as rf:
    for line in rf.readlines():
        img_name, img_label, img_index = line.strip().split()
        img_name = img_name[:27] + "annotations" + img_name[33:-3] + "png"
        img_path_list_tr.append(img_name)
        img_label_list_tr.append(int(img_label))

Class_num = 1040
all_cls_attributes_info = np.zeros((Class_num, 150))
img_label_list_tr = np.array(img_label_list_tr)
for i_cls in range(Class_num):
    print(i_cls)
    idx = np.where(img_label_list_tr == i_cls)[0]
    cur_img_path_list_tr = [img_path_list_tr[i] for i in idx]
    cls_attributes_info = np.zeros((len(cur_img_path_list_tr), 150))

    if len(cur_img_path_list_tr) > 0:
        for i_img in range(len(cur_img_path_list_tr)):
            image = misc.imread(cur_img_path_list_tr[i_img])
            image = image.flatten().tolist()
            objectList = list(set(image))
            objectList.sort()
            objectList = np.array(objectList)
            objectList = objectList[objectList != 0]
            cls_attributes_info[i_img, objectList-1] = 1
        cls_attributes_info = np.sum(cls_attributes_info, axis=0) / len(cur_img_path_list_tr)
        # cls_attributes_info = softmax(cls_attributes_info)
        all_cls_attributes_info[i_cls, :] = cls_attributes_info

# np.save('./ade/all_cls_attributes_info.npy', all_cls_attributes_info)
# all_cls_attributes_info = np.load('./ade/all_cls_attributes_info.npy')


def compute_ClsSimilarity_underObject(class_attributes):
    ClsSimilarity_underObject = np.zeros((1040, 1040))
    for i in range(1040):
        for j in range(i+1, 1040):
            if class_attributes[i] > 0 and class_attributes[j] > 0:
                ClsSimilarity_underObject[i, j] = (class_attributes[i] + class_attributes[j]) / 2.0
                ClsSimilarity_underObject[j, i] = ClsSimilarity_underObject[i, j]
    return ClsSimilarity_underObject

Object_num = 150
all_cls_attributes_info[all_cls_attributes_info < 0.3] = 0
similarityMatrix_cls_part = np.zeros((Class_num, Class_num, Object_num))
for i_obj in range(Object_num):
    print(i_obj)
    similarityMatrix_cls_part[:, :, i_obj] = compute_ClsSimilarity_underObject(all_cls_attributes_info[:, i_obj])

# np.save('./ade/similarityMatrix_cls_part.npy', similarityMatrix_cls_part)
# similarityMatrix_cls_part = np.load('./ade/similarityMatrix_cls_part.npy')

# threshold and remain most similar parts between each class pair
threshold = np.sort(similarityMatrix_cls_part.flatten())[int(0.99*Class_num*Class_num*Object_num)]
similarityMatrix_cls_part_copy = np.copy(similarityMatrix_cls_part)
similarityMatrix_cls_part_copy[similarityMatrix_cls_part >= threshold] = 1
similarityMatrix_cls_part_copy[similarityMatrix_cls_part < threshold] = 0
similarityMatrix_cls_part_copy = similarityMatrix_cls_part_copy.astype(int)


com_extracted_attributes = np.zeros((Class_num, Class_num), dtype=object)
for i in range(Class_num):
    for j in range(i+1, Class_num):
        object_idx = np.argwhere(similarityMatrix_cls_part_copy[i, j, :] == 1).flatten()
        object_idx = object_idx + 1
        object_idx = object_idx.tolist()
        com_extracted_attributes[i, j] = object_idx
        com_extracted_attributes[j, i] = object_idx
np.save('./ade/com_extracted_attributes_001.npy', com_extracted_attributes)