-
Notifications
You must be signed in to change notification settings - Fork 0
/
OfflineEncoder.py
40 lines (28 loc) · 1.1 KB
/
OfflineEncoder.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
#!/usr/bin/python3
from os.path import join;
from math import ceil;
import numpy as np;
import cv2;
import tensorflow as tf;
class Encoder(object):
def __init__(self, model_path = 'models'):
self.model = tf.keras.models.load_model(join(model_path, 'vggface2.h5'), compile = False);
def preprocess(self, img):
assert img.shape[2] == 3;
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB);
inputs = tf.expand_dims(img, axis = 0);
outputs = tf.image.resize_with_pad(inputs, 224, 224);
return outputs;
def batch(self, imgs):
inputs = [self.preprocess(img) for img in imgs];
outputs = tf.concat(inputs, axis = 0);
return outputs;
def encode(self, imgs):
assert type(imgs) is list;
if len(imgs) == 0: return tf.zeros((0,self.model.outputs[0].shape[-1]), dtype = tf.float32);
assert np.all([type(img) is np.ndarray and len(img.shape) == 3 for img in imgs]);
batch = self.batch(imgs);
return self.model(batch);
if __name__ == "__main__":
assert tf.executing_eagerly() == True;
encoder = Encoder();