Bird/test.py at main · Programming-in-AI/Bird · 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
import torch
import torchvision.models as models
import torchvision.transforms as transforms
from PIL import Image
from tqdm import tqdm
import numpy as np
from dataloader import CustomDataset, Dataloader, read_txt
import pandas as pd

def load_model(model_path):
    model = models.resnet101(weights='ResNet101_Weights.DEFAULT')
    model.fc = torch.nn.Linear(model.fc.in_features, 200)

    loaded_parameter = torch.load(model_path)
    model.load_state_dict(loaded_parameter, strict=False)
    return model


def test(model_path, input_img_path):

    # load model
    model = load_model(model_path)

    # make input have batch channel
    tr = transforms.ToTensor()
    input_img = tr(Image.open(input_img_path))
    input_img = torch.unsqueeze(input_img, 0)

    # model
    with torch.no_grad():
        model.eval()
        output = model(input_img)

    # output should be tensor type
    class_num = torch.argmax(output).item()
    print(f'This image is {class_num}th class')

    return class_num


def test_per_class(root_dir, model_path, device):
    print('[Test dataset Processing...]')
    dataset = CustomDataset(root_dir, isTrain=True)
    print("Training data size : {}".format(dataset.__len__()[0]))
    print("Validating data size : {}".format(dataset.__len__()[1]))
    batch_size = 32
    val_dataloader = Dataloader(dataset.val_dataset, batch_size)

    model = load_model(model_path)
    class_acc = np.zeros(200, dtype=float)
    right = np.zeros(200, dtype=int)
    wrong = np.zeros(200, dtype=int)
    wrong_index = np.array([])

    for i, (img, label) in tqdm(enumerate(val_dataloader), total=len(val_dataloader)):
        model = model.to(device)
        img = img.to(device)
        label = label.to(device)

        with torch.no_grad():
            h = model(img)
            _, y_pred = h.max(1)

        ans_list = (label == y_pred).float()

        for j in range(len(ans_list)):
            if ans_list[j] == 1:
                right[label[j]] += 1
            else:
                wrong[label[j]] += 1
                np.append(wrong_index, i*len(label) + j)

    # accuracy per class
    for i in range(len(class_acc)):
        class_acc[i] = round(right[i] / (right[i] + wrong[i]), 2)

    # make dataframe and save in txt
    df = pd.DataFrame({'class': [], 'acc': []})
    for i, item in enumerate(class_acc):
        df.loc[i] = [str(i) + 'th', str(item) + '%']
    df.to_csv('acc_per_class.txt', sep='\t', index=False)

    line = read_txt('./CUB_200_2011/train_test_split.txt')

    val_sequence = np.array([int(item.split(' ')[0]) for item in line if line.split(' ')[1] == '0'])

    return class_acc, val_sequence[wrong_index]


def test_per_class_topk(root_dir, model_path, device, top_k):
    print('[Test dataset Processing...]')
    dataset = CustomDataset(root_dir, isTrain=True)
    print("Training data size : {}".format(dataset.__len__()[0]))
    print("Validating data size : {}".format(dataset.__len__()[1]))
    batch_size = 16
    val_dataloader = Dataloader(dataset.val_dataset, batch_size)

    model = load_model(model_path)
    class_acc = np.zeros(200, dtype=float)
    right = np.zeros(200, dtype=int)
    wrong = np.zeros(200, dtype=int)
    wrong_index = np.array([])
    for i, (img, label) in tqdm(enumerate(val_dataloader), total=len(val_dataloader)):
        model = model.to(device)
        img = img.to(device)
        label = label.to(device)

        with torch.no_grad():
            h = model(img)
            _, y_pred = torch.topk(h, k=top_k, dim=1)

        ans_list = ([1 if label[i] in y_pred[i] else 0 for i in range(len(y_pred))])

        for j in range(len(ans_list)):
            if ans_list[j] == 1:
                right[label[j]] += 1
            else:
                wrong[label[j]] += 1
                np.append(wrong_index, i * len(label) + j)

    # accuracy per class
    for i in range(len(class_acc)):
        class_acc[i] = round(right[i] / (right[i] + wrong[i]) * 100, 2)

    # make dataframe and save in txt
    df = pd.DataFrame({'class': [], 'acc': []})
    for i, item in enumerate(class_acc):
        df.loc[i] = [str(i) + 'th', str(item) + '%']
    df.to_csv('topk_'+str(top_k)+'_acc_per_class.txt', sep='\t', index=False)

    line = read_txt('./CUB_200_2011/train_test_split.txt')

    val_sequence = np.array([int(item.split(' ')[0]) for item in line if line.split(' ')[1] == '0'])

    return class_acc, val_sequence[wrong_index]