forked from potterhsu/easy-faster-rcnn.pytorch
-
Notifications
You must be signed in to change notification settings - Fork 0
/
infer.py
60 lines (43 loc) · 2.23 KB
/
infer.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
import argparse
import random
from PIL import ImageDraw
from torchvision.transforms import transforms
from backbone.interface import Interface
from bbox import BBox
from dataset import Dataset
from model import Model
def _infer(path_to_input_image: str, path_to_output_image: str, path_to_checkpoint: str, backbone_name: str):
image = transforms.Image.open(path_to_input_image)
image_tensor, scale = Dataset.preprocess(image)
backbone = Interface.from_name(backbone_name)(pretrained=False)
model = Model(backbone).cuda()
model.load(path_to_checkpoint)
forward_input = Model.ForwardInput.Eval(image_tensor.cuda())
forward_output: Model.ForwardOutput.Eval = model.eval().forward(forward_input)
detection_bboxes = forward_output.detection_bboxes / scale
detection_labels = forward_output.detection_labels
detection_probs = forward_output.detection_probs
draw = ImageDraw.Draw(image)
for bbox, label, prob in zip(detection_bboxes.tolist(), detection_labels.tolist(), detection_probs.tolist()):
if prob < 0.6:
continue
color = random.choice(['red', 'green', 'blue', 'yellow', 'purple', 'white'])
bbox = BBox(left=bbox[0], top=bbox[1], right=bbox[2], bottom=bbox[3])
category = Dataset.LABEL_TO_CATEGORY_DICT[label]
draw.rectangle(((bbox.left, bbox.top), (bbox.right, bbox.bottom)), outline=color)
draw.text((bbox.left, bbox.top), text=f'{category:s} {prob:.3f}', fill=color)
image.save(path_to_output_image)
if __name__ == '__main__':
def main():
parser = argparse.ArgumentParser()
parser.add_argument('input', type=str, help='path to input image')
parser.add_argument('output', type=str, help='path to output result image')
parser.add_argument('-c', '--checkpoint', help='path to checkpoint')
parser.add_argument('-b', '--backbone', choices=['vgg16', 'resnet101'], required=True, help='name of backbone model')
args = parser.parse_args()
path_to_input_image = args.input
path_to_output_image = args.output
path_to_checkpoint = args.checkpoint
backbone_name = args.backbone
_infer(path_to_input_image, path_to_output_image, path_to_checkpoint, backbone_name)
main()