TPPI-Net/TPPP_predict.py at main · haochench/TPPI-Net · 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
"""
Evaluating inference time (whole HSI) and classification accuracy (test set) of TPPP-Nets
"""
import os
import torch
import argparse
import numpy as np
import yaml
import scipy.io as sio
import time
import auxil
from TPPI.utils import convert_state_dict
from TPPI.models import get_model
from sklearn.preprocessing import MinMaxScaler, StandardScaler


def predict_patches(data, model, cfg, device):
    transfer_data_start = time.time()
    data = data.to(device)
    transfer_data_end = time.time()
    transfer_time = transfer_data_end - transfer_data_start
    predicted = []
    bs = cfg["prediction"]["batch_size"]
    tsp = time.time()
    with torch.no_grad():
        for i in range(0, data.shape[0], bs):
            end_index = i + bs
            batch_data = data[i:end_index]
            outputs = model(batch_data)
            [predicted.append(a) for a in outputs.data.cpu().numpy()]
    tep = time.time()
    prediction_time = tep - tsp
    return prediction_time, transfer_time, np.array(predicted)


def timeCost_TPPP(cfg, logdir):
    name = cfg["data"]["dataset"]
    device = auxil.get_device()
    # Setup image
    teposition_path = 'dataset/split_dataset/testSet_position.npy'
    position = np.load(teposition_path)
    org_img_path = 'dataset/'
    if name == "IP":
        img = sio.loadmat(os.path.join(org_img_path, 'indian_pines_corrected.mat'))['indian_pines_corrected']
        gt = sio.loadmat(os.path.join(org_img_path, 'indian_pines_gt.mat'))['indian_pines_gt']
    elif name == "PU":
        img = sio.loadmat(os.path.join(org_img_path, 'paviaU.mat'))['paviaU']
        gt = sio.loadmat(os.path.join(org_img_path, 'paviaU_gt.mat'))['paviaU_gt']
    elif name == "SV":
        img = sio.loadmat(os.path.join(org_img_path, 'salinas_corrected.mat'))['salinas_corrected']
        gt = sio.loadmat(os.path.join(org_img_path, 'salinas_gt.mat'))['salinas_gt']
    else:
        print("No this dataset")
    print("data shape:", img.shape)
    print("GT shape:", gt.shape)

    # image processing
    time_pre_start = time.time()
    # StandardScaler
    shapeor = img.shape
    img = img.reshape(-1, img.shape[-1])
    img = StandardScaler().fit_transform(img)
    img = img.reshape(shapeor)
    # create patch
    data = auxil.creat_PP(cfg["data"]["PPsize"], img)
    # NHWC -> NCHW
    data = data.transpose(0, 3, 1, 2)
    data = torch.from_numpy(data).float()
    time_pre_end = time.time()
    time_pre_processing = time_pre_end - time_pre_start
    print("creat patch {} data over!", data.shape)

    # setup model:
    model = get_model(cfg['model'], cfg['data']['dataset'])
    state = convert_state_dict(
        torch.load(os.path.join(logdir, cfg["train"]["best_model_path"]))[
            "model_state"])
    model.load_state_dict(state)
    model.eval()
    model.to(device)

    # transfer model to GPU
    ts1 = time.time()
    model.to(device)
    ts2 = time.time()

    # predicting
    print("predicting...")
    pt, tt, outputs = predict_patches(data, model, cfg, device)
    print(outputs.shape)

    # get result and reshape
    comb_s = time.time()
    outputs = np.array(outputs)
    pred = np.argmax(outputs, axis=1)
    if cfg['data']['dataset'] == 'IP':
        pred = np.reshape(pred, (145, 145))
    elif cfg['data']['dataset'] == 'PU':
        pred = np.reshape(pred, (610, 340))
    elif cfg['data']['dataset'] == 'SV':
        pred = np.reshape(pred, (512, 217))
    elif cfg['data']['dataset'] == 'KSC':
        pred = np.reshape(pred, (512, 614))
    comb_e = time.time()

    # show predicted result
    pred += 1
    auxil.decode_segmap(pred)

    # computing classification accuracy
    pred = pred[position == 1]
    gt = gt[position == 1]
    classification, confusion, result = auxil.reports(pred, gt)
    result_info = "OA AA Kappa and each Acc:\n" + str(result)

    # report time cost and accuracy
    print("******************** Time ***********************")
    print("Pre_processing time is:", time_pre_processing)
    print("Transfer time is:", tt + (ts2-ts1), "  model:", ts2 - ts1, "  data:", tt)
    print("Prediction time is:", pt)
    print("combine time is:", comb_e-comb_s)
    print('Total inference time is:', time_pre_processing + tt + (ts2-ts1) +pt +comb_e-comb_s)

    # report classification accuracy
    print("****************** Accuracy *********************")
    print(result_info)
    print("\n")


if __name__ == "__main__":
    parser = argparse.ArgumentParser(description='PyTorch DCNNs Training')
    parser.add_argument(
        "--config",
        nargs="?",
        type=str,
        default="configs/config.yml",
        help="Configuration file to use",
    )
    args = parser.parse_args()
    with open(args.config) as fp:
        cfg = yaml.load(fp, Loader=yaml.FullLoader)
    logdir = os.path.join("runs", cfg["model"], str(cfg["run_ID"]))
    timeCost_TPPP(cfg, logdir)