GIRAFFEHD/eval.py at main · AustinXY/GIRAFFEHD · 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
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
import os
import sys
import argparse
import torch
from torchvision import utils
from model import GIRAFFEHDGenerator

def get_interval(args):
    if args.control_i == 4:
        if args.range_u is None:
            p0 = torch.tensor([[args.ckpt_args.range_u[0]]]).repeat(
                args.batch, 1).to(args.device)
            p1 = torch.tensor([[args.ckpt_args.range_u[1]]]).repeat(
                args.batch, 1).to(args.device)
        else:
            p0 = torch.tensor([[args.range_u[0]]]).repeat(
                args.batch, 1).to(args.device)
            p1 = torch.tensor([[args.range_u[1]]]).repeat(
                args.batch, 1).to(args.device)

    elif args.control_i == 5:
        if args.range_v is None:
            p0 = torch.tensor([[args.ckpt_args.range_v[0]]]).repeat(
                args.batch, 1).to(args.device)
            p1 = torch.tensor([[args.ckpt_args.range_v[1]]]).repeat(
                args.batch, 1).to(args.device)
        else:
            p0 = torch.tensor([[args.range_v[0]]]).repeat(
                args.batch, 1).to(args.device)
            p1 = torch.tensor([[args.range_v[1]]]).repeat(
                args.batch, 1).to(args.device)

    elif args.control_i == 6:
        print('Changing radius not implemented')
        sys.exit()

    elif args.control_i == 7:
        if args.scale_range_min is None:
            p0 = torch.tensor([args.ckpt_args.scale_range_min]).repeat(
                args.batch, 1).to(args.device)
            p1 = torch.tensor([args.ckpt_args.scale_range_max]).repeat(
                args.batch, 1).to(args.device)
        else:
            p0 = torch.tensor([args.scale_range_min]).repeat(
                args.batch, 1).to(args.device)
            p1 = torch.tensor([args.scale_range_max]).repeat(
                args.batch, 1).to(args.device)

    elif args.control_i == 8:
        if args.translation_range_min is None:
            p0 = torch.tensor([args.ckpt_args.translation_range_min]).repeat(
                args.batch, 1).to(args.device)
            p1 = torch.tensor([args.ckpt_args.translation_range_max]).repeat(
                args.batch, 1).to(args.device)
        else:
            p0 = torch.tensor([args.translation_range_min]).repeat(
                args.batch, 1).to(args.device)
            p1 = torch.tensor([args.translation_range_max]).repeat(
                args.batch, 1).to(args.device)

    elif args.control_i == 9:
        if args.rotation_range is None:
            p0 = torch.tensor([[args.ckpt_args.rotation_range[0]]]).repeat(
                args.batch, 1).to(args.device)
            p1 = torch.tensor([[args.ckpt_args.rotation_range[1]]]).repeat(
                args.batch, 1).to(args.device)
        else:
            p0 = torch.tensor([[args.rotation_range[0]]]).repeat(
                args.batch, 1).to(args.device)
            p1 = torch.tensor([[args.rotation_range[1]]]).repeat(
                args.batch, 1).to(args.device)

#     elif args.control_i == 10:
#         if args.bg_translation_range_min is None:
#             p0 = torch.tensor([args.ckpt_args.bg_translation_range_min]).repeat(
#                 args.batch, 1).to(args.device)
#             p1 = torch.tensor([args.ckpt_args.bg_translation_range_max]).repeat(
#                 args.batch, 1).to(args.device)
#         else:
#             p0 = torch.tensor([args.bg_translation_range_min]).repeat(
#                 args.batch, 1).to(args.device)
#             p1 = torch.tensor([args.bg_translation_range_max]).repeat(
#                 args.batch, 1).to(args.device)

#     elif args.control_i == 11:
#         if args.bg_rotation_range is None:
#             p0 = torch.tensor([[args.ckpt_args.bg_rotation_range[0]]]).repeat(
#                 args.batch, 1).to(args.device)
#             p1 = torch.tensor([[args.ckpt_args.bg_rotation_range[1]]]).repeat(
#                 args.batch, 1).to(args.device)
#         else:
#             p0 = torch.tensor([[args.bg_rotation_range[0]]]).repeat(
#                 args.batch, 1).to(args.device)
#             p1 = torch.tensor([[args.bg_rotation_range[1]]]).repeat(
#                 args.batch, 1).to(args.device)

    return p0, p1


def eval(args, generator):
    generator.eval()

    img_li = []
    if args.control_i in list(range(0,4)):
        img_rep = generator.get_rand_rep(args.batch)
        for i in range(args.n_sample):
            _img_rep = generator.get_rand_rep(args.batch)
            img_rep[args.control_i] = _img_rep[args.control_i]
            img = generator(img_rep=img_rep, inject_index=args.inj_idx, mode='eval')[0]
            img_li.append(img)

    if args.control_i in list(range(4,10)):
        p0, p1 = get_interval(args)
        delta = (p1 - p0) / (args.n_sample - 1)

        img_rep = generator.get_rand_rep(args.batch)
        for i in range(args.n_sample):
            p = p0 + delta * i
            img_rep[args.control_i] = p
            img = generator(img_rep=img_rep, inject_index=args.inj_idx, mode='eval')[0]
            img_li.append(img)

    img = []
    for b in range(args.batch):
        b_li = []
        for i in range(args.n_sample):
            b_li.append(img_li[i][b:b+1])

        img.append(torch.cat(b_li))
    img = torch.cat(img)

    utils.save_image(
        img,
        f"eval/control_{args.control_i}.png",
        nrow=args.n_sample,
        normalize=True,
        range=(-1, 1),
    )


if __name__ == "__main__":
    device = "cuda"

    parser = argparse.ArgumentParser(description="Giraffe trainer")

    parser.add_argument('--ckpt', type=str, default=None, help='path to the checkpoint')

    parser.add_argument('--batch', type=int, default=16, help='batch size')
    parser.add_argument('--n_sample', type=int, default=8, help='number of the samples generated')
    parser.add_argument('--inj_idx', type=int, default=-1, help='inject index for evaluation')

    parser.add_argument("--control_i", type=int, default=0, help='control index')
    # 0: fg_shape; 1: fg_app; 2: bg_shape;    3: bg_app;   4: camera rotation angle;  5: elevation angle;
    # --: radius;  7: scale;  8: translation; 9: rotation;

    args = parser.parse_args()

    n_gpu = int(os.environ["WORLD_SIZE"]) if "WORLD_SIZE" in os.environ else 1
    args.device = device

    assert args.ckpt is not None
    print("load model:", args.ckpt)

    ckpt = torch.load(args.ckpt, map_location=lambda storage, loc: storage)
    args.ckpt_args = ckpt['args']

    # change interpolation ranges if needed
    args.scale_range_min = None  # [0.2, 0.16, 0.16]
    args.scale_range_max = None  # [0.25, 0.2, 0.2]

    args.translation_range_min = None  # [-0.22, -0.12, -0.06]
    args.translation_range_max = None  # [0.22, 0.12, 0.08]

    args.rotation_range = None  # [0., 1.]

    args.bg_translation_range_min = None  # [-0.2, -0.2, 0.]
    args.bg_translation_range_max = None  # [0.2, 0.2, 0.]

    args.bg_rotation_range = None  # [0., 0.]

    args.range_u = None  # [0., 0.]

    args.range_v = None  # [0.41667, 0.5]


    if args.inj_idx == -1:
        if args.ckpt_args.size == 256:
            args.inj_idx = 2
        elif args.ckpt_args.size == 512:
            args.inj_idx = 4
        elif args.ckpt_args.size == 1024:
            args.inj_idx = 4

    generator = GIRAFFEHDGenerator(
        device=device,
        z_dim=args.ckpt_args.z_dim,
        z_dim_bg=args.ckpt_args.z_dim_bg,
        size=args.ckpt_args.size,
        resolution_vol=args.ckpt_args.res_vol,
        feat_dim=args.ckpt_args.feat_dim,
        range_u=args.ckpt_args.range_u,
        range_v=args.ckpt_args.range_v,
        fov=args.ckpt_args.fov,
        scale_range_max=args.ckpt_args.scale_range_max,
        scale_range_min=args.ckpt_args.scale_range_min,
        translation_range_max=args.ckpt_args.translation_range_max,
        translation_range_min=args.ckpt_args.translation_range_min,
        rotation_range=args.ckpt_args.rotation_range,
        bg_translation_range_max=args.ckpt_args.bg_translation_range_max,
        bg_translation_range_min=args.ckpt_args.bg_translation_range_min,
        bg_rotation_range=args.ckpt_args.bg_rotation_range,
        refine_n_styledconv=2,
        refine_kernal_size=3,
        grf_use_mlp=args.ckpt_args.grf_use_mlp,
        pos_share=args.ckpt_args.pos_share,
        use_viewdirs=args.ckpt_args.use_viewdirs,
        grf_use_z_app=args.ckpt_args.grf_use_z_app,
        fg_gen_mask=args.ckpt_args.fg_gen_mask
    ).to(device)

    generator.load_state_dict(ckpt["g_ema"])

    with torch.no_grad():
        eval(args, generator)