LogicMD/train.py at main · beforeint/LogicMD · 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
import copy
import torch
import numpy as np
from torch.utils.data import DataLoader
from sklearn.metrics import accuracy_score, confusion_matrix
# from tqdm import tqdm
# from dataset import FeatureDataset
from model import SimilarityModule, DetectionModule


# Configs
DEVICE = "cuda:0"
NUM_WORKER = 1
BATCH_SIZE = 64
LR = 1e-3
L2 = 0  # 1e-5
NUM_EPOCH = 100


def prepare_data(text, image, label):
    nr_index = [i for i, l in enumerate(label) if l == 1]
    text_nr = text[nr_index]
    image_nr = image[nr_index]
    fixed_text = copy.deepcopy(text_nr)
    matched_image = copy.deepcopy(image_nr)
    unmatched_image = copy.deepcopy(image_nr).roll(shifts=3, dims=0)
    return fixed_text, matched_image, unmatched_image


def train():
    # ---  Load Config  ---


    # ---  Build Model & Trainer  ---
    similarity_module = SimilarityModule()
    detection_module = DetectionModule()

    loss_func_similarity = torch.nn.CosineEmbeddingLoss()
    loss_func_detection = torch.nn.CrossEntropyLoss()
    # ---  Model Training  ---
    loss_similarity_total = 0
    loss_detection_total = 0
    best_acc = 0


def test(similarity_module, detection_module, test_loader):
    similarity_module.eval()
    detection_module.eval()

    device = torch.device(DEVICE)
    loss_func_detection = torch.nn.CrossEntropyLoss()
    loss_func_similarity = torch.nn.CosineEmbeddingLoss()

    similarity_count = 0
    detection_count = 0
    loss_similarity_total = 0
    loss_detection_total = 0
    similarity_label_all = []
    detection_label_all = []
    similarity_pre_label_all = []
    detection_pre_label_all = []

    with torch.no_grad():
        for i, (text, image, label) in enumerate(test_loader):
            batch_size = text.shape[0]
            text = text.to(device)
            image = image.to(device)
            label = label.to(device)

            fixed_text, matched_image, unmatched_image = prepare_data(text, image, label)
            fixed_text.to(device)
            matched_image.to(device)
            unmatched_image.to(device)

            # ---  TASK1 Similarity  ---

            text_aligned_match, image_aligned_match, pred_similarity_match = similarity_module(fixed_text, matched_image)
            text_aligned_unmatch, image_aligned_unmatch, pred_similarity_unmatch = similarity_module(fixed_text, unmatched_image)
            similarity_pred = torch.cat([pred_similarity_match.argmax(1), pred_similarity_unmatch.argmax(1)], dim=0)
            similarity_label_0 = torch.cat([torch.ones(pred_similarity_match.shape[0]), torch.zeros(pred_similarity_unmatch.shape[0])], dim=0).to(device)
            similarity_label_1 = torch.cat([torch.ones(pred_similarity_match.shape[0]), -1 * torch.ones(pred_similarity_unmatch.shape[0])], dim=0).to(device)

            text_aligned_4_task1 = torch.cat([text_aligned_match, text_aligned_unmatch], dim=0)
            image_aligned_4_task1 = torch.cat([image_aligned_match, image_aligned_unmatch], dim=0)
            loss_similarity = loss_func_similarity(text_aligned_4_task1, image_aligned_4_task1, similarity_label_1)

            # ---  TASK2 Detection  ---

            text_aligned, image_aligned, _ = similarity_module(text, image)
            pre_detection = detection_module(text, image, text_aligned, image_aligned)
            loss_detection = loss_func_detection(pre_detection, label)
            pre_label_detection = pre_detection.argmax(1)

            # ---  Record  ---

            loss_similarity_total += loss_similarity.item() * (2 * fixed_text.shape[0])
            loss_detection_total += loss_detection.item() * text.shape[0]
            similarity_count += (fixed_text.shape[0] * 2)
            detection_count += text.shape[0]

            similarity_pre_label_all.append(similarity_pred.detach().cpu().numpy())
            detection_pre_label_all.append(pre_label_detection.detach().cpu().numpy())
            similarity_label_all.append(similarity_label_0.detach().cpu().numpy())
            detection_label_all.append(label.detach().cpu().numpy())

        loss_similarity_test = loss_similarity_total / similarity_count
        loss_detection_test = loss_detection_total / detection_count

        similarity_pre_label_all = np.concatenate(similarity_pre_label_all, 0)
        detection_pre_label_all = np.concatenate(detection_pre_label_all, 0)
        similarity_label_all = np.concatenate(similarity_label_all, 0)
        detection_label_all = np.concatenate(detection_label_all, 0)

        acc_similarity_test = accuracy_score(similarity_pre_label_all, similarity_label_all)
        acc_detection_test = accuracy_score(detection_pre_label_all, detection_label_all)
        cm_similarity = confusion_matrix(similarity_pre_label_all, similarity_label_all)
        cm_detection = confusion_matrix(detection_pre_label_all, detection_label_all)

    return acc_similarity_test, acc_detection_test, loss_similarity_test, loss_detection_test, cm_similarity, cm_detection


if __name__ == "__main__":
    similarity_module = SimilarityModule()
    detection_module = DetectionModule()
    print("parameters of similarity_module", sum(p.numel() for p in similarity_module.parameters() if p.requires_grad))
    print("parameters of detection_module", sum(p.numel() for p in detection_module.parameters() if p.requires_grad))