-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathrun_gradio.py
193 lines (140 loc) · 6.73 KB
/
run_gradio.py
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
import cv2
import torch
import os, glob
import numpy as np
import gradio as gr
from PIL import Image, ImageFilter
from omegaconf import OmegaConf
from contextlib import nullcontext
from pytorch_lightning import seed_everything
from os.path import join as ospj
from util import *
def predict(cfgs, model, sampler, batch):
context = torch.no_grad
with context():
batch, batch_uc_1 = prepare_batch(cfgs, batch)
c, uc_1 = model.conditioner.get_unconditional_conditioning(
batch,
batch_uc=batch_uc_1,
force_uc_zero_embeddings=cfgs.force_uc_zero_embeddings,
)
x = sampler.get_init_noise(cfgs, batch=batch)
samples_z = sampler(model, x, cond=c, batch=batch, uc=uc_1, init_step=0,
detailed = cfgs.detailed)
samples_x = model.decode_first_stage(samples_z)
samples = torch.clamp((samples_x + 1.0) / 2.0, min=0.0, max=1.0)
return samples, samples_z
def demo_predict(input_blk, text, num_samples, steps, scale, seed, show_detail):
global cfgs, global_index
global_index += 1
if num_samples > 1: cfgs.noise_iters = 0
cfgs.batch_size = num_samples
cfgs.steps = steps
cfgs.scale[0] = scale
cfgs.detailed = show_detail
seed_everything(seed)
sampler = init_sampling(cfgs)
image = input_blk["image"]
mask = input_blk["mask"]
image = cv2.resize(image, (cfgs.W, cfgs.H))
mask = cv2.resize(mask, (cfgs.W, cfgs.H))
mask = (mask == 0).astype(np.int32)
image = torch.from_numpy(image.transpose(2,0,1)).to(dtype=torch.float32) / 127.5 - 1.0
mask = torch.from_numpy(mask.transpose(2,0,1)).to(dtype=torch.float32).mean(dim=0, keepdim=True)
masked = image * mask
mask = 1 - mask
seg_mask = torch.cat((torch.ones(len(text)), torch.zeros(cfgs.seq_len-len(text))))
# additional cond
txt = f"\"{text}\""
original_size_as_tuple = torch.tensor((cfgs.H, cfgs.W))
crop_coords_top_left = torch.tensor((0, 0))
target_size_as_tuple = torch.tensor((cfgs.H, cfgs.W))
image = torch.tile(image[None], (num_samples, 1, 1, 1))
mask = torch.tile(mask[None], (num_samples, 1, 1, 1))
masked = torch.tile(masked[None], (num_samples, 1, 1, 1))
seg_mask = torch.tile(seg_mask[None], (num_samples, 1))
original_size_as_tuple = torch.tile(original_size_as_tuple[None], (num_samples, 1))
crop_coords_top_left = torch.tile(crop_coords_top_left[None], (num_samples, 1))
target_size_as_tuple = torch.tile(target_size_as_tuple[None], (num_samples, 1))
text = [text for i in range(num_samples)]
txt = [txt for i in range(num_samples)]
name = [str(global_index) for i in range(num_samples)]
batch = {
"image": image,
"mask": mask,
"masked": masked,
"seg_mask": seg_mask,
"label": text,
"txt": txt,
"original_size_as_tuple": original_size_as_tuple,
"crop_coords_top_left": crop_coords_top_left,
"target_size_as_tuple": target_size_as_tuple,
"name": name
}
samples, samples_z = predict(cfgs, model, sampler, batch)
samples = samples.cpu().numpy().transpose(0, 2, 3, 1) * 255
results = [Image.fromarray(sample.astype(np.uint8)) for sample in samples]
if cfgs.detailed:
sections = []
attn_map = Image.open(f"./temp/attn_map/mask_{global_index}.png")
seg_maps = np.load(f"./temp/seg_map/seg_{global_index}.npy")
for i, seg_map in enumerate(seg_maps):
seg_map = cv2.resize(seg_map, (cfgs.W, cfgs.H))
sections.append((seg_map, text[0][i]))
seg = (results[0], sections)
else:
attn_map = None
seg = None
return results, attn_map, seg
if __name__ == "__main__":
os.system(f"rm -rf ./temp")
os.makedirs("./temp", exist_ok=True)
os.makedirs("./temp/attn_map", exist_ok=True)
os.makedirs("./temp/seg_map", exist_ok=True)
cfgs = OmegaConf.load("./configs/demo.yaml")
model = init_model(cfgs)
global_index = 0
block = gr.Blocks().queue()
with block:
with gr.Row():
gr.HTML(
"""
<div style="text-align: center; max-width: 1200px; margin: 20px auto;">
<h1 style="font-weight: 600; font-size: 2rem; margin: 0.5rem;">
High Fidelity Scene Text Synthesis
</h1>
<div align=center><img src="file/demo/teaser.png" alt="DreamText" width="80%"></div>
</div>
"""
)
with gr.Row():
with gr.Column():
input_blk = gr.Image(source='upload', tool='sketch', type="numpy", label="Input", height=512)
text = gr.Textbox(label="Text to render:", info="the text you want to render at the masked region")
run_button = gr.Button(variant="primary")
with gr.Accordion("Advanced options", open=True):
num_samples = gr.Slider(label="Images", info="number of generated images, locked as 1", minimum=1, maximum=1, value=1, step=1)
steps = gr.Slider(label="Steps", info ="denoising sampling steps", minimum=1, maximum=200, value=50, step=1)
scale = gr.Slider(label="Guidance Scale", info="the scale of classifier-free guidance (CFG)", minimum=0.0, maximum=10.0, value=2.0, step=0.1)
seed = gr.Slider(label="Seed", info="random seed for noise initialization", minimum=0, maximum=2147483647, step=1, randomize=True)
show_detail = gr.Checkbox(label="Show Detail", info="show the additional visualization results", value=True)
with gr.Column():
gallery = gr.Gallery(label="Output", height=512, preview=True)
with gr.Accordion("Visualization results", open=True):
with gr.Tab(label="Latent Character Masks"):
attn_map = gr.Image(show_label=False, show_download_button=True)
with gr.Tab(label="Attention Rendering"):
seg_map = gr.AnnotatedImage(height=384, show_label=True)
# examples
examples = []
example_paths = sorted(glob.glob(ospj("./demo/examples", "*")))
for example_path in example_paths:
label = example_path.split(os.sep)[-1].split(".")[0].split("_")[0]
examples.append([example_path, label])
gr.Markdown("## Examples:")
gr.Examples(
examples=examples,
inputs=[input_blk, text]
)
run_button.click(fn=demo_predict, inputs=[input_blk, text, num_samples, steps, scale, seed, show_detail], outputs=[gallery, attn_map, seg_map])
block.launch()