-
Notifications
You must be signed in to change notification settings - Fork 0
/
rtree_query_manager.py
38 lines (32 loc) · 1.34 KB
/
rtree_query_manager.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
import pickle
from typing import Tuple, List
from rtree import index
import face_recognition
import numpy as np
from helpers import measure_execution_time
class RTreeQueryManager:
@measure_execution_time
def __init__(self, m: int, collection: List[Tuple[str, np.ndarray]]) -> None:
p = index.Property()
p.dimension = 128 # D
p.buffering_capacity = m # M
self.collection_ = collection
self.idx = index.Index(properties=p)
for i in range(len(collection)):
self.idx.insert(id=i, coordinates=collection[i][1])
@measure_execution_time
def knn_query(self, q: str, k: int) -> List[List[Tuple[str, float]]]:
image_query = face_recognition.load_image_file(q)
query_embeds = face_recognition.face_encodings(image_query)
total_result = list()
for face_embed in query_embeds:
nearest = self.idx.nearest(coordinates=face_embed, num_results=k)
partial_result = list()
for item_id in nearest:
obj = self.collection_[item_id]
partial_result.append((obj[0], np.linalg.norm(obj[1] - face_embed)))
total_result.append(partial_result)
return total_result
if __name__ == "__main__":
with open("out.embeds", mode="rb") as collection_file:
collection = pickle.load(collection_file)