-
Notifications
You must be signed in to change notification settings - Fork 0
/
evaluators.py
131 lines (110 loc) · 4.9 KB
/
evaluators.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
import sys
import logging
import numpy as np
from typing import Tuple
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
class Evaluator:
def __init__(self, num_samples: int = 0, num_features: int = 0):
self.loss = 0.0
self.best_loss = sys.maxsize
self.best_image2text_recall_at_k = (-1.0, -1.0, -1.0)
self.cur_image2text_recall_at_k = (-1.0, -1.0, -1.0)
self.best_text2image_recall_at_k = (-1.0, -1.0, -1.0)
self.cur_text2image_recall_at_k = (-1.0, -1.0, -1.0)
self.index_update = 0
self.num_samples = num_samples
self.num_features = num_features
self.embedded_images = np.zeros((self.num_samples, self.num_features))
self.embedded_captions = np.zeros((self.num_samples, self.num_features))
def reset_all_vars(self) -> None:
self.loss = 0
self.index_update = 0
self.embedded_images = np.zeros((self.num_samples, self.num_features))
self.embedded_captions = np.zeros((self.num_samples, self.num_features))
self.cur_text2image_recall_at_k = -1.0
self.cur_image2text_recall_at_k = -1.0
def update_metrics(self, loss: float) -> None:
self.loss += loss
def update_embeddings(
self, embedded_images: np.ndarray, embedded_captions: np.ndarray
) -> None:
num_samples = embedded_images.shape[0]
self.embedded_images[
self.index_update : self.index_update + num_samples, :
] = embedded_images
self.embedded_captions[
self.index_update : self.index_update + num_samples, :
] = embedded_captions
self.index_update += num_samples
def is_best_loss(self) -> bool:
if self.loss < self.best_loss:
return True
return False
def update_best_loss(self):
self.best_loss = self.loss
def is_best_recall_at_k(self) -> bool:
# Update current
self.cur_image2text_recall_at_k = self.image2text_recall_at_k()
self.cur_text2image_recall_at_k = self.text2image_recall_at_k()
# Sum recalls
image2text_recall_at_ks = sum(self.cur_image2text_recall_at_k)
text2image_recall_at_ks = sum(self.cur_text2image_recall_at_k)
# Sum best recalls
best_image2text_recall_at_ks = sum(self.best_image2text_recall_at_k)
best_text2image_recall_at_ks = sum(self.best_text2image_recall_at_k)
# Check if the current are the better
if (image2text_recall_at_ks + text2image_recall_at_ks) > (
best_image2text_recall_at_ks + best_text2image_recall_at_ks
):
return True
return False
def update_best_recall_at_k(self):
self.best_image2text_recall_at_k = self.cur_image2text_recall_at_k
self.best_text2image_recall_at_k = self.cur_text2image_recall_at_k
def image2text_recall_at_k(self) -> Tuple[float, float, float]:
"""Computes the recall at K when doing image to text retrieval and updates the
object variable.
Returns:
The recall at 1, 5, 10.
"""
num_images = self.embedded_images.shape[0] // 5
ranks = np.zeros(num_images)
for index in range(num_images):
# Get query image
query_image = self.embedded_images[5 * index]
# Similarities
similarities = np.dot(query_image, self.embedded_captions.T).flatten()
indices = np.argsort(similarities)[::-1]
# Score
rank = sys.maxsize
for i in range(5 * index, 5 * index + 5, 1):
tmp = np.where(indices == i)[0][0]
if tmp < rank:
rank = tmp
ranks[index] = rank
r1 = 100.0 * len(np.where(ranks < 1)[0]) / len(ranks)
r5 = 100.0 * len(np.where(ranks < 5)[0]) / len(ranks)
r10 = 100.0 * len(np.where(ranks < 10)[0]) / len(ranks)
return r1, r5, r10
def text2image_recall_at_k(self) -> Tuple[float, float, float]:
"""Computes the recall at K when doing text to image retrieval and updates the
object variable.
Returns:
The recall at 1, 5, 10.
"""
num_captions = self.embedded_captions.shape[0]
ranks = np.zeros(num_captions)
for index in range(num_captions):
# Get query captions
query_captions = self.embedded_captions[5 * index : 5 * index + 5]
# Similarities
similarities = np.dot(query_captions, self.embedded_images[0::5].T)
inds = np.zeros(similarities.shape)
for i in range(len(inds)):
inds[i] = np.argsort(similarities[i])[::-1]
ranks[5 * index + i] = np.where(inds[i] == index)[0][0]
r1 = 100.0 * len(np.where(ranks < 1)[0]) / len(ranks)
r5 = 100.0 * len(np.where(ranks < 5)[0]) / len(ranks)
r10 = 100.0 * len(np.where(ranks < 10)[0]) / len(ranks)
return r1, r5, r10