DataDetective/LossBased.py at main · SSCT-Lab/DataDetective · 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
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
# smooth L1 Loss computation
import json
import math
import time

import torch
import torch.nn.functional as F
from matplotlib import pyplot as plt
import random
from UTILS.parameters import parameters
from UTILS.metric import Metric

from UTILS.FocalLoss import FocalLoss

fault_type_dict = parameters().fault_type
params = parameters()
metric = Metric()

# convert fault_type_dict to number2fault
number2fault = {}
for key in fault_type_dict.keys():
    number2fault[fault_type_dict[key]] = key


class LossBased:
    def __init__(self,
                 config={"dataset": "VOC", "model": "frcnn", "fault_ratio": 0.1, "is_dirty": True, "set": "train",
                         "loss_type": "ce"},
                 missing_dict=None,excel=None):
        print('LossBased\n')
        self.excel=excel
        self.config = config
        self.missing_dict = missing_dict
        self.gt_path = './data/fault_annotations/' + self.config["dataset"] + self.config['set'] + '_mixedfault' + str(
            self.config["fault_ratio"]) + '.json'
        det_name_ = lambda d, m, f, x: m + 'dirty' + str(
            f) + '_' + d + self.config['set'] + '_inferences.json' if x else m + 'clean_' + d + '_inferences.json'
        self.det_path = './data/detection_results/' + det_name_(self.config["dataset"], self.config["model"],
                                                                self.config["fault_ratio"], self.config["is_dirty"])

        self.losstype = self.config['loss_type']
        print(self.gt_path, self.det_path)
        with open(self.gt_path, 'r') as f:
            self.gt = json.load(f)
        with open(self.det_path, 'r') as f:
            self.det = json.load(f)
        self.det = [i for i in self.det if i["score"] > params.m_t]

        fault_num = {
            'no fault': 0,
            'class fault': 0,
            'location fault': 0,
            'redundancy fault': 0,
            'missing fault': 0,
            'findable missing fault': 0,
        }
        with open(self.gt_path, 'r') as f:
            gt = json.load(f)
        for i in gt:
            fault_num[number2fault[i["fault_type"]]] += 1

        self.fault_num = fault_num

        # transform dec to {imagename:[]} format dict
        self.dec_dict = {}
        for i in range(len(self.det)):
            if self.det[i]["image_name"] in self.dec_dict:
                self.dec_dict[self.det[i]["image_name"]].append(self.det[i])
            else:
                self.dec_dict[self.det[i]["image_name"]] = [self.det[i]]
        self.imagename2boxes = {}
        for instance in self.gt:
            if instance["fault_type"] != fault_type_dict['missing fault']:
                if instance["image_name"] not in self.imagename2boxes:
                    self.imagename2boxes[instance["image_name"]] = []
                self.imagename2boxes[instance["image_name"]].append(instance["boxes"])

    def run(self):
        start_time = time.time()
        results = []
        for i in range(len(self.gt)):
            # print progress bar
            print('\r', 'progress: ', i, '/', len(self.gt), end='')

            if self.gt[i]['fault_type'] != fault_type_dict['missing fault'] and self.gt[i][
                "image_name"] in self.dec_dict:
                # loss = min loss of all decs
                min_loss = 100000

                # decious=[self.dec_dict[self.gt[i]["image_name"]][j]['bbox'] for j in range(len(self.dec_dict[self.gt[i]["image_name"]]))]
                #
                # IoUs = metric.cal_IoU([self.gt[i]["boxes"]], decious)
                #
                # max_iou_index = torch.argmax(IoUs, dim=1).item()
                #
                # min_loss = self.compute_loss(self.dec_dict[self.gt[i]["image_name"]][max_iou_index]['full_scores'],
                #                                 self.gt[i]['labels'],
                #                                 self.dec_dict[self.gt[i]["image_name"]][max_iou_index]['bbox'],
                #                                 self.gt[i]['boxes'])

                for j in range(len(self.dec_dict[self.gt[i]["image_name"]])):

                    loss = self.compute_loss(self.dec_dict[self.gt[i]["image_name"]][j]["full_scores"],
                                             self.gt[i]["labels"],
                                             self.dec_dict[self.gt[i]["image_name"]][j]["bbox"],
                                             self.gt[i]["boxes"], self.losstype)
                    if loss < min_loss:
                        min_loss = loss

                results.append({"loss": min_loss, "fault_type": self.gt[i]["fault_type"], 'detectiongt_category_id': -1,
                                'image_name': self.gt[i]["image_name"]})

        # add len(self.dec_dict) to results

        # random shuffle self.gt_image_names
        self.gt_image_names = [i for i in self.imagename2boxes.keys()]
        random.shuffle(self.gt_image_names)

        for name in self.gt_image_names:
            if name in self.missing_dict:
                results.append({"loss": 0, "fault_type": fault_type_dict['missing fault'], 'detectiongt_category_id': 0,
                                'image_name': name})
            else:
                results.append({"loss": 0, "fault_type": fault_type_dict['no fault'], 'detectiongt_category_id': -1,
                                'image_name': name})

        # sort results by loss from large to small
        results.sort(key=lambda x: x["loss"], reverse=True)
        end_time = time.time()
        print(self.losstype + " loss time: ", end_time - start_time)
        # X = [i for i in range(len(results))]
        # Y = [0 for i in range(len(results))]
        # # print(results)
        # fault_t = []
        #
        # flag_list = []
        print(metric.APFD(results))
        EXAM_F, EXAM_F_rel, Top_1, Top_3 = metric.EXAM_F(results)

        col_offset = None
        if self.config['loss_type'] == 'ce':
            col_offset = 2
        elif self.config['loss_type'] == 'focal':
            col_offset = 3
        self.excel.run([EXAM_F_rel, EXAM_F, Top_1, Top_3], [0, 12, 24, 36], col_offset)
        print('lossbased EXAM_F: ', EXAM_F)
        print('lossbased EXAM_F_rel: ', EXAM_F_rel)
        print('lossbased Top_1: ', Top_1)
        print('lossbased Top_3: ', Top_3)
        # for i in range(len(results)):
        #
        #     if results[i]['fault_type'] != fault_type_dict['no fault']:
        #         Y[i] = Y[i - 1] + 1
        #         fault_t.append(results[i]['fault_type'])
        #     else:
        #         Y[i] = Y[i - 1]
        #         fault_t.append(fault_type_dict["no fault"])
        #
        # # different color for different fault type
        # color_dict = {0: 'b', 1: 'g', 2: 'r', 3: 'c', 4: 'm', 5: 'y', 6: 'k', 7: 'w'}
        #
        # # delete no fault
        # for i in range(len(fault_t)):
        #     if fault_t[i] == fault_type_dict["no fault"]:
        #         fault_t[i] = -1
        # XX = [X[i] for i in range(len(X)) if fault_t[i] != -1]
        # YY = [Y[i] for i in range(len(Y)) if fault_t[i] != -1]
        # fault_t = [i for i in fault_t if i != -1]
        #
        # plt.plot(X, Y, color='b')
        # # plt.scatter(XX, YY, c=[color_dict[i] for i in fault_t])
        # # plt.plot([0, len(results)], [0, len(fault_t)], color='r')
        # plt.show()
        # # color legend
        # for i in range(len(fault_type_dict)):
        #     plt.scatter([], [], c=color_dict[i], label=number2fault[i])

        return results
        # plt.legend()
        #
        # plt.show()

        # print('\n')
        # metric.RAUC(results, self.fault_num, rauc_num=params.rauc_num)
        #
        # print(len(results))
        # # get top params.vocfrcnn_FaultSet_length results
        # if self.detective_cofig == 'vocfrcnn':
        #     results = results[:params.vocfrcnn_FaultSet_length]
        #     fault_ratio, fault_inclusiveness = metric.RateAndInclusiveness(results, self.fault_num)
        #     print('fault_ratio: ', fault_ratio)
        #     print('fault_inclusiveness: ', fault_inclusiveness)

    def compute_loss(self, full_scores, label, box_pre, box_gt, loss_type):

        if loss_type == 'ce':
            return F.cross_entropy(torch.tensor(full_scores), torch.tensor(label)).item() \
                + F.smooth_l1_loss(torch.tensor(box_pre), torch.tensor(box_gt)).item()

        if loss_type == 'focal':
            return FocalLoss(gamma=2)(torch.tensor(full_scores), torch.tensor(label)).item() \
                + F.smooth_l1_loss(torch.tensor(box_pre), torch.tensor(box_gt)).item()


if __name__ == "__main__":
    loss_based = LossBased()
    loss_based.run()