robot-memory-visualization/instance_classifier.py at main · Mersen130/robot-memory-visualization · GitHub

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
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
import os
from collections import defaultdict
import torch
import torchvision.transforms as transforms
import torchvision.models as models
import numpy as np

from utilities import parse_box

class YoloClass:
    def __init__(self, clsid2name, ) -> None:
        pass

__location__ = os.path.realpath(
    os.path.join(os.getcwd(), os.path.dirname(__file__)))

class InstanceClassifier:
    def __init__(self) -> None:
        self.model = models.mobilenet_v2(pretrained=True)
        self.model = torch.nn.Sequential(*list(self.model.children())[0])
        self.model.eval()
        self.model.to("cuda")

        self.transform = transforms.Compose([
            transforms.ToPILImage(),
            transforms.Resize(256),
            transforms.CenterCrop(224),
            transforms.ToTensor(),
            transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
        ])

        self.clsid2name = {}
        self.clsid2concreteID = {}
        self.concreteid2embed = {}
        with open(os.path.join(__location__, 'yolo_classes.txt'), 'r') as file:
            lines = file.readlines()
            for index, line in enumerate(lines):
                name = line.strip()
                self.clsid2name[index] = name
                self.clsid2concreteID[index] = []
        # self.name2clsid = {v: k for k, v in self.clsid2name.items()}

        self.next_id = len(self.clsid2name)

    def get_embedding(self, input_batch):

        if torch.cuda.is_available():
            input_batch = input_batch.to('cuda')

        embeddings = None
        with torch.no_grad():
            embeddings = self.model(input_batch)

        # Flatten the embeddings
        embeddings = torch.flatten(embeddings, start_dim=1)

        return embeddings.cpu()

    def cosine_similarity(self, vector1, vector2):
        dot_product = np.dot(vector1, vector2)
        norm1 = np.linalg.norm(vector1)
        norm2 = np.linalg.norm(vector2)
        similarity = dot_product / (norm1 * norm2)
        return similarity

    def update(self, frame, boxes):
        curr_objs, curr_obj2cls, curr_obj2box = parse_box(boxes)

        same_class_objs = defaultdict(list)
        for key, value in curr_obj2cls.items():
            same_class_objs[value].append(key)

        obj_mapping = {}
        for clsid, objs in same_class_objs.items():
            if clsid == 0:
                # don't check human
                continue

            # get cropped images
            cropped_images = []
            for obj in objs:
                box = curr_obj2box[obj]
                if box[0] < 0:
                    box[0] = 0
                if box[1] < 0:
                    box[1] = 0
                if box[2] < 0:
                    box[2] = 0
                if box[3] < 0:
                    box[3] = 0

                cropped_images.append(self.transform(frame[int(box[1]):int(box[3]), int(box[0]):int(box[2])]))
            cropped_images = torch.stack(cropped_images)

            embeddings = self.get_embedding(cropped_images)

            # print(clsid)
            if self.clsid2concreteID[clsid]:  # this class has multiple instances
                for i in range(len(objs)):
                    subject = embeddings[i]
                    for j in range(len(self.clsid2concreteID[clsid])):
                        target = self.concreteid2embed[self.clsid2concreteID[clsid][j]]
                        simi = self.cosine_similarity(subject, target)
                        # print(simi)
                        if simi > 0.5:
                            obj_mapping[objs[i]] = self.clsid2concreteID[clsid][j]
                            break  # TODO: alternative: choose the highest similarity

                    if objs[i] not in obj_mapping:
                        self.clsid2concreteID[clsid].append(self.next_id)  # TODO: cache to local yolo_classes.txt
                        self.clsid2name[self.next_id] = self.clsid2name[clsid] + str(len(self.clsid2concreteID[clsid]))
                        self.concreteid2embed[self.next_id] = subject
                        obj_mapping[objs[i]] = self.next_id
                        self.next_id += 1
            else:
                for i in range(len(objs)):
                    subject = embeddings[i]

                    self.clsid2concreteID[clsid].append(self.next_id)  # cache to local yolo_classes.txt
                    self.clsid2name[self.next_id] = self.clsid2name[clsid] + str(len(self.clsid2concreteID[clsid]))
                    self.concreteid2embed[self.next_id] = subject
                    obj_mapping[objs[i]] = self.next_id
                    self.next_id += 1

            # print(self.clsid2concreteID)
            # print(self.clsid2name)
            # print(obj_mapping)

        # modify class ids in boxes
        boxes = self.apply_mapping(boxes, obj_mapping)
        # print(boxes)
        return boxes, self.clsid2name, {v: k for k, v in self.clsid2name.items()}


    def apply_mapping(self, boxes, obj_mapping):
        for i in range(len(boxes)):
            if boxes[i][5] != 0:
                obj_id = int(boxes[i][4])
                boxes[i][5] = obj_mapping[obj_id]
        return boxes


instance_classifier = InstanceClassifier()