[视觉工程]以图搜图之搜索策略(bf,kdTree,ballTree,annoy,nms,falconn)

import numpy as np
import pickle
from tqdm import tqdm, tqdm_notebook
import random
import time
from sklearn.manifold import TSNE
from sklearn.decomposition import PCA
import PIL
from PIL import Image
from sklearn.neighbors import NearestNeighbors

import glob
import matplotlib
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
%matplotlib inline

filenames = pickle.load(open('data/filenames-caltech101.pickle', 'rb'))
feature_list = pickle.load(open('data/features-caltech101-resnet.pickle', 'rb'))
class_ids = pickle.load(open('data/class_ids-caltech101.pickle', 'rb'))

num_images = len(filenames)
num_features_per_image = len(feature_list[0])
print("Number of images = ", num_images)
print("Number of features per image = ", num_features_per_image)

random_image_index = random.randint(0, num_images)

# Standard features + Brute Force Algorithm on one image
%timeit NearestNeighbors(n_neighbors=5, algorithm='brute', metric='euclidean').fit(feature_list)
neighbors = NearestNeighbors(n_neighbors=5,
                             algorithm='brute',
                             metric='euclidean').fit(feature_list)
%timeit neighbors.kneighbors([feature_list[random_image_index]])

# Standard features + k-d Tree Algorithm on one image
%timeit NearestNeighbors(n_neighbors=5, algorithm='kd_tree').fit(feature_list)
neighbors = NearestNeighbors(n_neighbors=5,
                             algorithm='kd_tree').fit(feature_list)

%timeit neighbors.kneighbors([feature_list[random_image_index]])

# Standard features + Ball Tree Algorithm on one image 
%timeit NearestNeighbors(n_neighbors=5, algorithm='ball_tree').fit(feature_list)
neighbors = NearestNeighbors(n_neighbors=5,
                             algorithm='ball_tree').fit(feature_list)

%timeit