forked from alvinwan/tricking-neural-networks
/
step_2_pretrained.py
51 lines (42 loc) · 1.63 KB
/
step_2_pretrained.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
"""Run ImageNet-pretrained ResNet18 on arbitrary image"""
from PIL import Image
import json
import torchvision.models as models
import torchvision.transforms as transforms
import torch
import sys
def get_idx_to_label():
with open("assets/imagenet_idx_to_label.json") as f:
return json.load(f)
def get_image_transform():
transform = transforms.Compose([
transforms.Resize(224), # resize smaller side of image to 224
transforms.CenterCrop(224), # take center 224x224 crop
transforms.ToTensor(), # convert from image object to PyTorch tensor, which our PyTorch model needs
transforms.Normalize(mean=[0.485, 0.456, 0.406], # normalize images, according to https://pytorch.org/docs/stable/torchvision/models.html
std=[0.229, 0.224, 0.225])
])
return transform
def load_image():
assert len(sys.argv) > 1, 'Need to pass path to image'
image = Image.open(sys.argv[1])
# transform image into correct format
transform = get_image_transform()
image = transform(image)[None]
return image
def predict(image):
# load pretrained ResNet18 model
model = models.resnet18(pretrained=True)
model.eval() # set model in 'evaluation' mode
# evaluate model on image
out = model(image)
# translate output into human-readable text
_, pred = torch.max(out, 1) # get class with highest probability
idx_to_label = get_idx_to_label() # get mapping from index to name
cls = idx_to_label[str(int(pred))] # get name, using index
return cls
def main():
x = load_image()
print(f'Prediction: {predict(x)}')
if __name__ == '__main__':
main()