AnyFashionAttributeEditing/inference_pairpca_editing.py at main · ArtmeScienceLab/AnyFashionAttributeEditing · 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
import torch
import os
from models.psp_hfgi import pSp
import torchvision.transforms as transforms
from PIL import Image
import argparse
import warnings
from tqdm import tqdm
import numpy as np
import time
warnings.filterwarnings("ignore")


def to_numpy(data):
    """Converts the input data to `numpy.ndarray`."""
    if isinstance(data, (int, float)):
        return np.array(data)
    if isinstance(data, np.ndarray):
        return data
    if isinstance(data, torch.Tensor):
        return data.detach().cpu().numpy()
    raise TypeError(f'Not supported data type `{type(data)}` for '
                    f'converting to `numpy.ndarray`!')

def get_keys(d, name):
    if 'state_dict' in d:
        d = d['state_dict']
    d_filt = {k[len(name) + 1:]: v for k, v in d.items() if k[:len(name)] == name}
    return d_filt

img_loadfold = transforms.Compose([
			transforms.Resize((256, 256)),
			transforms.ToTensor(),
			transforms.Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5])])

def tensor2im(var):
	# var shape: (3, H, W)
	var = var.cpu().detach().transpose(0, 2).transpose(0, 1).numpy()
	var = ((var + 1) / 2)
	var[var < 0] = 0
	var[var > 1] = 1
	var = var * 255
	return Image.fromarray(var.astype('uint8'))


def get_convert(net, base_dir, save_dir, direction, rate):
    img_list = os.listdir(base_dir)
    device = 'cuda'
    wr = torch.from_numpy(np.load(direction)).to(device)
    generator = net.decoder
    generator.eval()

    # times = []
    for i in tqdm(range(len(img_list))):
        img_path = os.path.join(base_dir, img_list[i])
        save_path = os.path.join(save_dir, img_list[i])
        np_name = img_list[i][:-4] + '.pt'
        save_w_path = os.path.join(save_dir, np_name)

        from_im = Image.open(img_path).convert('RGB')
        from_im = img_loadfold(from_im)
        from_im = from_im.unsqueeze(0).to(device)

        code = net.encoder(from_im)
        latent_code = code + net.latent_avg.repeat(code.shape[0], 1, 1)

        # print(save_w_path)
        # torch.save(latent_code.detach().cpu(), save_w_path)

        wr_add  = latent_code + rate * wr
        edit, _ = generator(wr_add ,input_is_latent=True)
        edit_resize = torch.nn.functional.interpolate(torch.clamp(edit, -1., 1.), size=(256,256) , mode='bilinear')

        # torch.save(w.detach().cpu(), save_w_path)
        rec, _ = generator(latent_code ,input_is_latent=True)
        rec_resize = torch.nn.functional.interpolate(torch.clamp( rec, -1., 1.), size=(256,256) , mode='bilinear')
        rec_x = (from_im - rec_resize).detach()
        res_edit_aligned = net.grid_align(torch.cat((rec_x, edit_resize), 1))
        rec_res = torch.nn.functional.interpolate(torch.clamp(res_edit_aligned, -1., 1.), size=(256,128) , mode='bilinear')
        res_conditions = net.residue(rec_res)

        if res_conditions is not None:
            add_edit_images, result_latent = net.decoder(wr_add, res_conditions, input_is_latent=True)
            # print(add_edit_images.size())
            img = add_edit_images.squeeze()
            # exit()
        result = tensor2im(img)
        result.save(save_path)
        # time_end = time.time()


    # print(np.mean(times))


def setup_model(checkpoint_path, device='cuda'):
    ckpt = torch.load(checkpoint_path, map_location='cpu')
    # print('load ckpt: ',checkpoint_path)
    opts = ckpt['opts']
    net = pSp(opts)
    net.load_state_dict(ckpt['state_dict'], strict=True)
    if 'latent_avg' in ckpt:
        net.latent_avg = ckpt['latent_avg'].to(device)
    # print(net)
    net.eval()
    net = net.to(device)
    return net, opts


def main():

    device = "cuda"
    ckpt  = '/path_to/ckpt/fp/8_twinnet_h_product.pt' # path to the checkpoint
    net, opts = setup_model(ckpt, device)

    base_dir = '' # base directory
    save_dir = '' # save directory
    direction = '' # pair PCA direction
    rate = '' # editing rate, rate=0 means no editing only reconstruction


    get_convert(net, base_dir, save_dir, direction, rate)
    return


if __name__ == '__main__':
    main()