Spaces:
Running
Running
Refactor
Browse files
app.py
CHANGED
@@ -3,20 +3,11 @@
|
|
3 |
from __future__ import annotations
|
4 |
|
5 |
import argparse
|
6 |
-
import os
|
7 |
-
import pickle
|
8 |
-
import sys
|
9 |
|
10 |
import gradio as gr
|
11 |
-
import numpy as np
|
12 |
-
import torch
|
13 |
-
import torch.nn as nn
|
14 |
from huggingface_hub import hf_hub_download
|
15 |
|
16 |
-
|
17 |
-
|
18 |
-
TOKEN = os.environ['TOKEN']
|
19 |
-
|
20 |
|
21 |
def parse_args() -> argparse.Namespace:
|
22 |
parser = argparse.ArgumentParser()
|
@@ -30,76 +21,9 @@ def parse_args() -> argparse.Namespace:
|
|
30 |
return parser.parse_args()
|
31 |
|
32 |
|
33 |
-
class App:
|
34 |
-
|
35 |
-
def __init__(self, device: torch.device):
|
36 |
-
self.device = device
|
37 |
-
self.model = self.load_model('stylegan_human_v2_1024.pkl')
|
38 |
-
|
39 |
-
def load_model(self, file_name: str) -> nn.Module:
|
40 |
-
path = hf_hub_download('hysts/StyleGAN-Human',
|
41 |
-
f'models/{file_name}',
|
42 |
-
use_auth_token=TOKEN)
|
43 |
-
with open(path, 'rb') as f:
|
44 |
-
model = pickle.load(f)['G_ema']
|
45 |
-
model.eval()
|
46 |
-
model.to(self.device)
|
47 |
-
with torch.inference_mode():
|
48 |
-
z = torch.zeros((1, model.z_dim)).to(self.device)
|
49 |
-
label = torch.zeros([1, model.c_dim], device=self.device)
|
50 |
-
model(z, label, force_fp32=True)
|
51 |
-
return model
|
52 |
-
|
53 |
-
def generate_z(self, z_dim: int, seed: int) -> torch.Tensor:
|
54 |
-
return torch.from_numpy(np.random.RandomState(seed).randn(
|
55 |
-
1, z_dim)).to(self.device).float()
|
56 |
-
|
57 |
-
@torch.inference_mode()
|
58 |
-
def generate_single_image(self, seed: int,
|
59 |
-
truncation_psi: float) -> np.ndarray:
|
60 |
-
seed = int(np.clip(seed, 0, np.iinfo(np.uint32).max))
|
61 |
-
|
62 |
-
z = self.generate_z(self.model.z_dim, seed)
|
63 |
-
label = torch.zeros([1, self.model.c_dim], device=self.device)
|
64 |
-
|
65 |
-
out = self.model(z,
|
66 |
-
label,
|
67 |
-
truncation_psi=truncation_psi,
|
68 |
-
force_fp32=True)
|
69 |
-
out = (out.permute(0, 2, 3, 1) * 127.5 + 128).clamp(0, 255).to(
|
70 |
-
torch.uint8)
|
71 |
-
return out[0].cpu().numpy()
|
72 |
-
|
73 |
-
@torch.inference_mode()
|
74 |
-
def generate_interpolated_images(
|
75 |
-
self, seed0: int, psi0: float, seed1: int, psi1: float,
|
76 |
-
num_intermediate: int) -> list[np.ndarray]:
|
77 |
-
seed0 = int(np.clip(seed0, 0, np.iinfo(np.uint32).max))
|
78 |
-
seed1 = int(np.clip(seed1, 0, np.iinfo(np.uint32).max))
|
79 |
-
|
80 |
-
z0 = self.generate_z(self.model.z_dim, seed0)
|
81 |
-
z1 = self.generate_z(self.model.z_dim, seed1)
|
82 |
-
vec = z1 - z0
|
83 |
-
dvec = vec / (num_intermediate + 1)
|
84 |
-
zs = [z0 + dvec * i for i in range(num_intermediate + 2)]
|
85 |
-
dpsi = (psi1 - psi0) / (num_intermediate + 1)
|
86 |
-
psis = [psi0 + dpsi * i for i in range(num_intermediate + 2)]
|
87 |
-
|
88 |
-
label = torch.zeros([1, self.model.c_dim], device=self.device)
|
89 |
-
|
90 |
-
res = []
|
91 |
-
for z, psi in zip(zs, psis):
|
92 |
-
out = self.model(z, label, truncation_psi=psi, force_fp32=True)
|
93 |
-
out = (out.permute(0, 2, 3, 1) * 127.5 + 128).clamp(0, 255).to(
|
94 |
-
torch.uint8)
|
95 |
-
out = out[0].cpu().numpy()
|
96 |
-
res.append(out)
|
97 |
-
return res
|
98 |
-
|
99 |
-
|
100 |
def main():
|
101 |
args = parse_args()
|
102 |
-
app =
|
103 |
|
104 |
with gr.Blocks(theme=args.theme) as demo:
|
105 |
gr.Markdown('''<center><h1>StyleGAN-Human</h1></center>
|
|
|
3 |
from __future__ import annotations
|
4 |
|
5 |
import argparse
|
|
|
|
|
|
|
6 |
|
7 |
import gradio as gr
|
|
|
|
|
|
|
8 |
from huggingface_hub import hf_hub_download
|
9 |
|
10 |
+
from model import Model
|
|
|
|
|
|
|
11 |
|
12 |
def parse_args() -> argparse.Namespace:
|
13 |
parser = argparse.ArgumentParser()
|
|
|
21 |
return parser.parse_args()
|
22 |
|
23 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
def main():
|
25 |
args = parse_args()
|
26 |
+
app = Model(device=args.device)
|
27 |
|
28 |
with gr.Blocks(theme=args.theme) as demo:
|
29 |
gr.Markdown('''<center><h1>StyleGAN-Human</h1></center>
|
model.py
ADDED
@@ -0,0 +1,81 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from __future__ import annotations
|
2 |
+
|
3 |
+
import os
|
4 |
+
import pickle
|
5 |
+
import sys
|
6 |
+
|
7 |
+
import numpy as np
|
8 |
+
import torch
|
9 |
+
import torch.nn as nn
|
10 |
+
from huggingface_hub import hf_hub_download
|
11 |
+
|
12 |
+
sys.path.insert(0, 'StyleGAN-Human')
|
13 |
+
|
14 |
+
HF_TOKEN = os.environ['HF_TOKEN']
|
15 |
+
|
16 |
+
|
17 |
+
class Model:
|
18 |
+
|
19 |
+
def __init__(self, device: str | torch.device):
|
20 |
+
self.device = torch.device(device)
|
21 |
+
self.model = self.load_model('stylegan_human_v2_1024.pkl')
|
22 |
+
|
23 |
+
def load_model(self, file_name: str) -> nn.Module:
|
24 |
+
path = hf_hub_download('hysts/StyleGAN-Human',
|
25 |
+
f'models/{file_name}',
|
26 |
+
use_auth_token=HF_TOKEN)
|
27 |
+
with open(path, 'rb') as f:
|
28 |
+
model = pickle.load(f)['G_ema']
|
29 |
+
model.eval()
|
30 |
+
model.to(self.device)
|
31 |
+
with torch.inference_mode():
|
32 |
+
z = torch.zeros((1, model.z_dim)).to(self.device)
|
33 |
+
label = torch.zeros([1, model.c_dim], device=self.device)
|
34 |
+
model(z, label, force_fp32=True)
|
35 |
+
return model
|
36 |
+
|
37 |
+
def generate_z(self, z_dim: int, seed: int) -> torch.Tensor:
|
38 |
+
return torch.from_numpy(np.random.RandomState(seed).randn(
|
39 |
+
1, z_dim)).to(self.device).float()
|
40 |
+
|
41 |
+
@torch.inference_mode()
|
42 |
+
def generate_single_image(self, seed: int,
|
43 |
+
truncation_psi: float) -> np.ndarray:
|
44 |
+
seed = int(np.clip(seed, 0, np.iinfo(np.uint32).max))
|
45 |
+
|
46 |
+
z = self.generate_z(self.model.z_dim, seed)
|
47 |
+
label = torch.zeros([1, self.model.c_dim], device=self.device)
|
48 |
+
|
49 |
+
out = self.model(z,
|
50 |
+
label,
|
51 |
+
truncation_psi=truncation_psi,
|
52 |
+
force_fp32=True)
|
53 |
+
out = (out.permute(0, 2, 3, 1) * 127.5 + 128).clamp(0, 255).to(
|
54 |
+
torch.uint8)
|
55 |
+
return out[0].cpu().numpy()
|
56 |
+
|
57 |
+
@torch.inference_mode()
|
58 |
+
def generate_interpolated_images(
|
59 |
+
self, seed0: int, psi0: float, seed1: int, psi1: float,
|
60 |
+
num_intermediate: int) -> list[np.ndarray]:
|
61 |
+
seed0 = int(np.clip(seed0, 0, np.iinfo(np.uint32).max))
|
62 |
+
seed1 = int(np.clip(seed1, 0, np.iinfo(np.uint32).max))
|
63 |
+
|
64 |
+
z0 = self.generate_z(self.model.z_dim, seed0)
|
65 |
+
z1 = self.generate_z(self.model.z_dim, seed1)
|
66 |
+
vec = z1 - z0
|
67 |
+
dvec = vec / (num_intermediate + 1)
|
68 |
+
zs = [z0 + dvec * i for i in range(num_intermediate + 2)]
|
69 |
+
dpsi = (psi1 - psi0) / (num_intermediate + 1)
|
70 |
+
psis = [psi0 + dpsi * i for i in range(num_intermediate + 2)]
|
71 |
+
|
72 |
+
label = torch.zeros([1, self.model.c_dim], device=self.device)
|
73 |
+
|
74 |
+
res = []
|
75 |
+
for z, psi in zip(zs, psis):
|
76 |
+
out = self.model(z, label, truncation_psi=psi, force_fp32=True)
|
77 |
+
out = (out.permute(0, 2, 3, 1) * 127.5 + 128).clamp(0, 255).to(
|
78 |
+
torch.uint8)
|
79 |
+
out = out[0].cpu().numpy()
|
80 |
+
res.append(out)
|
81 |
+
return res
|