def test_hash_distinct_images(self, hash_function):
# Put in distinct images and check that hamming distance between hashes is large
hash_im_1 = hash_function(PATH_SINGLE_IMAGE)
hash_im_2 = hash_function(p.parent /
'data/mixed_images/ukbench09268.jpg')
hamdist = Hashing.hamming_distance(hash_im_1, hash_im_2)
assert hamdist > 20
def test_hash_small_rotation(self, hash_function):
# Rotate image slightly (1 degree) and check that hamming distance between hashes is not too large
orig_image = Image.open(PATH_SINGLE_IMAGE)
rotated_image = np.array(orig_image.rotate(1))
hash_im_1 = hash_function(image_array=np.array(orig_image))
hash_im_2 = hash_function(image_array=rotated_image)
hamdist = Hashing.hamming_distance(hash_im_1, hash_im_2)
assert hamdist < 3
def hasher():
hashobj = Hashing()
return hashobj
def test_hash_resize(self, hash_function):
# Resize one image to (300, 300) and check that hamming distance between hashes is not too large
hash_im_1 = hash_function(PATH_SINGLE_IMAGE)
hash_im_2 = hash_function(PATH_SINGLE_IMAGE_RESIZED)
hamdist = Hashing.hamming_distance(hash_im_1, hash_im_2)
assert hamdist < 3
import sys
import pytest
import numpy as np
from sklearn.metrics.pairwise import cosine_similarity
from imagededup.handlers.search.retrieval import (
HashEval,
cosine_similarity_chunk,
get_cosine_similarity,
)
from imagededup.methods.hashing import Hashing
HAMMING_DISTANCE_FUNCTION = Hashing().hamming_distance
def test_cosine_similarity_chunk():
X = np.random.rand(333, 100)
start_idx = 10
end_idx = 100
input_tuple = (X, (start_idx, end_idx))
result = cosine_similarity_chunk(input_tuple)
np.testing.assert_array_almost_equal(
result, cosine_similarity(X[start_idx:end_idx, :], X).astype('float16')
)
def test_get_cosine_similarity():
X = np.random.rand(333, 10)