-
Notifications
You must be signed in to change notification settings - Fork 0
/
predictor.py
60 lines (49 loc) · 2.05 KB
/
predictor.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
# Predict bbox in image
# https://github.com/cr00z/virtual_tryon
import numpy as np
import torch
import matplotlib.pyplot as plt
from detectron2 import model_zoo
from detectron2.config import get_cfg
from detectron2.data import MetadataCatalog
from detectron2.engine import DefaultPredictor
from detectron2.utils.visualizer import Visualizer
detectron2_config = 'COCO-Detection/faster_rcnn_R_50_FPN_3x.yaml'
detectron2_weights = 'detectron2://COCO-Detection/faster_rcnn_R_50_FPN_3x/' \
'137849458/model_final_280758.pkl'
person_class = 0
class Predictor:
def __init__(self):
self.cfg = get_cfg()
self.cfg.merge_from_file(model_zoo.get_config_file(detectron2_config))
if not torch.cuda.is_available():
self.cfg.MODEL.DEVICE = 'cpu'
self.cfg.MODEL.ROI_HEADS.SCORE_THRESH_TEST = 0.5
self.cfg.MODEL.WEIGHTS = detectron2_weights
self.predictor = DefaultPredictor(self.cfg)
self.outputs = None
def predict(self, img):
self.outputs = self.predictor(img)
return self.outputs
def get_max_body_bbox(self):
""" Sort the person bbox using size and get max """
person_idxs = self.outputs['instances'].pred_classes == person_class
person_bboxes = self.outputs['instances'][person_idxs].pred_boxes.tensor.cpu()
bboxes_size = [(c[2] - c[0]) * (c[3] - c[1]) for c in person_bboxes]
max_body_bbox = person_bboxes[np.argmax(bboxes_size)]
# convert to XYWH
max_body_bbox[2] = max_body_bbox[2] - max_body_bbox[0]
max_body_bbox[3] = max_body_bbox[3] - max_body_bbox[1]
return max_body_bbox
# TODO: remove
def visualise(self, img):
""" Method for visualize some test results """
v = Visualizer(
img[:, :, ::-1],
MetadataCatalog.get(self.cfg.DATASETS.TRAIN[0]),
scale=1.2
)
v = v.draw_instance_predictions(self.outputs['instances'].to('cpu'))
plt.figure(figsize=(12, 8))
plt.imshow(v.get_image()[:, :, ::-1])
plt.show()