PPTFusion/run.py at main · thfylsty/PPTFusion · 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
import os
import argparse
import os
import argparse
from PIL import Image
import cv2
import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
import torchvision
import torch.utils.data
from torch.utils.data import DataLoader
from torchvision.utils import save_image
import torchvision.transforms as transforms
from util import *
from networks import *
from tqdm import tqdm
from ssim import SSIM

def parse_args():
    parser = argparse.ArgumentParser()

    # parser.add_argument('--images_path', default='../../data/vi', help='Path to drive image')
    # parser.add_argument('--images_path', default='/data/Disk_A/fuyu/code/data/ir_vi', help='Path to drive image')
    parser.add_argument('--images_path', default='/data/Disk_B/MSCOCO2014/train2014', help='Path to drive image')

    parser.add_argument('--batch_size', default=1, help='batch size')
    parser.add_argument('--eopch', default=1000, help='eopch')
    parser.add_argument('--learning_rate', default=0.0001, help='learning_rate')
    parser.add_argument('--load_weights', default=True, help='load weights')
    return parser.parse_args()

torch.backends.cudnn.enabled = False

if __name__ == "__main__":
    os.environ["CUDA_VISIBLE_DEVICES"] = "1"
    args = parse_args()
    print(args)
    startepoch = 20
    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
    dataset = DatasetFromFolder(args.images_path)
    dataloader = DataLoader(dataset, batch_size=args.batch_size,
                            shuffle=True, num_workers=4, drop_last=True)

    generator = Generator().to(device)
    # generator = torch.nn.DataParallel(generator)

    if args.load_weights:
        generator.load_state_dict(torch.load('./weights/G_eopch_last_l1.pth'))

    optimizer_gen = optim.Adam(generator.parameters(), lr=args.learning_rate)


    # optimizer_gen = torch.nn.DataParallel(optimizer_gen)
    MSE = nn.MSELoss()
    l1 = nn.L1Loss()
    ssim = SSIM()
    # vgg_loss = VGGLoss()
    loss_G=0
    datalen = len(dataloader)
    print(datalen)

    for epoch in range(startepoch,args.eopch):
        generator.train()
        data = tqdm(dataloader)
        for en, x in enumerate(data):
            x = x.to(device)
            # print(x.shape)
            # print(x, x.max(),x.mean())
            # img_ir = img_ir.to(device)
            fuse_re = generator(x)
            # print(fuse_re.shape,x.shape)
            optimizer_gen.zero_grad()

            l1loss = MSE(fuse_re, x)*10
            # l1loss = l1(fuse_re, x)
            ssim_loss = (ssim(fuse_re,x))
            ssim_loss = (1-ssim_loss)
            # print(fuse_image.shape)
            # loss_VGG = vgg_loss(fuse_image, img_vi,img_ir)
            # loss_G = l1loss+loss_VGG+ssim_loss
            loss_G = l1loss+ssim_loss

            loss_G.backward()
            optimizer_gen.step()
            # print("It %s:  Loss G:%.5f[%.5f %.5f %.5f] " % (
            # en, loss_G.item(), l1loss.item(), ssim_loss.item(), loss_VGG.item()))
            printstring = "epoch[%d][%d/%d] Loss G:%.5f [%.5f %.5f] " %(epoch,en+1, datalen,loss_G.item(),l1loss.item(),ssim_loss.item())
            data.set_description(printstring)

            if en%200==0:
                img = torchvision.utils.make_grid([x[0].cpu(),fuse_re[0].cpu()],nrow=2)
                save_image(img, fp=(os.path.join('output/img_train' + str(epoch) + '.jpg')))
            if en%10000==0:
                img = torchvision.utils.make_grid([x[0].cpu(),fuse_re[0].cpu()],nrow=2)
                save_image(img, fp=(os.path.join('output/img_train' + str(epoch) + '_en' + str(en) + '.jpg')))

        # with torch.no_grad():
        #     generator.eval()
        #     fake_test = generator(img_test_vi.unsqueeze(0),img_test_ir.unsqueeze(0))
        # save_image(fake_test, fp=(os.path.join('output/fake_img_test' + str(epoch) + '.jpg')))

        # print("Epoch %s:  Loss G:%.5f[%.5f %.5f %.5f] " %(epoch, loss_G.item(), l1loss.item(),ssim_loss.item(),loss_VGG.item()))

        if epoch%5==0:
            torch.save(generator,"weights/G_eopch_"+str(epoch)+".pth")
        if epoch % 1 == 0:
            torch.save(generator.state_dict(), "weights/G_eopch_last.pth")
    torch.save(generator,"weights/G_eopch_final.pth")