feat: add gradio demo
Browse files
demo.py
ADDED
@@ -0,0 +1,179 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import argparse
|
2 |
+
import os
|
3 |
+
import gradio as gr
|
4 |
+
from PIL import Image
|
5 |
+
from torchvision import transforms
|
6 |
+
from detector.model import *
|
7 |
+
from detector import config
|
8 |
+
from font_dataset.font import load_fonts, load_font_with_exclusion
|
9 |
+
|
10 |
+
parser = argparse.ArgumentParser()
|
11 |
+
parser.add_argument(
|
12 |
+
"-d",
|
13 |
+
"--device",
|
14 |
+
type=int,
|
15 |
+
default=0,
|
16 |
+
help="GPU devices to use (default: 0), -1 for CPU",
|
17 |
+
)
|
18 |
+
parser.add_argument(
|
19 |
+
"-c",
|
20 |
+
"--checkpoint",
|
21 |
+
type=str,
|
22 |
+
default=None,
|
23 |
+
help="Trainer checkpoint path (default: None)",
|
24 |
+
)
|
25 |
+
parser.add_argument(
|
26 |
+
"-m",
|
27 |
+
"--model",
|
28 |
+
type=str,
|
29 |
+
default="resnet18",
|
30 |
+
choices=["resnet18", "resnet34", "resnet50", "resnet101", "deepfont"],
|
31 |
+
help="Model to use (default: resnet18)",
|
32 |
+
)
|
33 |
+
parser.add_argument(
|
34 |
+
"-f",
|
35 |
+
"--font-classification-only",
|
36 |
+
action="store_true",
|
37 |
+
help="Font classification only (default: False)",
|
38 |
+
)
|
39 |
+
parser.add_argument(
|
40 |
+
"-z",
|
41 |
+
"--size",
|
42 |
+
type=int,
|
43 |
+
default=512,
|
44 |
+
help="Model feature image input size (default: 512)",
|
45 |
+
)
|
46 |
+
parser.add_argument(
|
47 |
+
"-s",
|
48 |
+
"--share",
|
49 |
+
action="store_true",
|
50 |
+
help="Get public link via Gradio (default: False)",
|
51 |
+
)
|
52 |
+
|
53 |
+
args = parser.parse_args()
|
54 |
+
|
55 |
+
config.INPUT_SIZE = args.size
|
56 |
+
device = torch.device("cpu") if args.device == -1 else torch.device("cuda", args.device)
|
57 |
+
|
58 |
+
regression_use_tanh = False
|
59 |
+
|
60 |
+
if args.model == "resnet18":
|
61 |
+
model = ResNet18Regressor(regression_use_tanh=regression_use_tanh)
|
62 |
+
elif args.model == "resnet34":
|
63 |
+
model = ResNet34Regressor(regression_use_tanh=regression_use_tanh)
|
64 |
+
elif args.model == "resnet50":
|
65 |
+
model = ResNet50Regressor(regression_use_tanh=regression_use_tanh)
|
66 |
+
elif args.model == "resnet101":
|
67 |
+
model = ResNet101Regressor(regression_use_tanh=regression_use_tanh)
|
68 |
+
elif args.model == "deepfont":
|
69 |
+
assert args.pretrained is False
|
70 |
+
assert args.size == 105
|
71 |
+
assert args.font_classification_only is True
|
72 |
+
model = DeepFontBaseline()
|
73 |
+
else:
|
74 |
+
raise NotImplementedError()
|
75 |
+
|
76 |
+
if torch.__version__ >= "2.0" and os.name == "posix":
|
77 |
+
model = torch.compile(model)
|
78 |
+
|
79 |
+
detector = FontDetector(
|
80 |
+
model=model,
|
81 |
+
lambda_font=1,
|
82 |
+
lambda_direction=1,
|
83 |
+
lambda_regression=1,
|
84 |
+
font_classification_only=args.font_classification_only,
|
85 |
+
lr=1,
|
86 |
+
betas=(1, 1),
|
87 |
+
num_warmup_iters=1,
|
88 |
+
num_iters=1e9,
|
89 |
+
num_epochs=1e9,
|
90 |
+
)
|
91 |
+
detector.load_from_checkpoint(
|
92 |
+
args.checkpoint,
|
93 |
+
map_location=device,
|
94 |
+
model=model,
|
95 |
+
lambda_font=1,
|
96 |
+
lambda_direction=1,
|
97 |
+
lambda_regression=1,
|
98 |
+
font_classification_only=args.font_classification_only,
|
99 |
+
lr=1,
|
100 |
+
betas=(1, 1),
|
101 |
+
num_warmup_iters=1,
|
102 |
+
num_iters=1e9,
|
103 |
+
num_epochs=1e9,
|
104 |
+
)
|
105 |
+
detector = detector.to(device)
|
106 |
+
detector.eval()
|
107 |
+
|
108 |
+
|
109 |
+
transform = transforms.Compose(
|
110 |
+
[
|
111 |
+
transforms.Resize((512, 512)),
|
112 |
+
transforms.ToTensor(),
|
113 |
+
]
|
114 |
+
)
|
115 |
+
|
116 |
+
print("Preparing fonts ...")
|
117 |
+
font_list, exclusion_rule = load_fonts()
|
118 |
+
|
119 |
+
font_list = list(filter(lambda x: not exclusion_rule(x), font_list))
|
120 |
+
font_list.sort(key=lambda x: x.path)
|
121 |
+
|
122 |
+
for i in range(len(font_list)):
|
123 |
+
font_list[i].path = font_list[i].path[18:] # remove ./dataset/fonts/./ prefix
|
124 |
+
|
125 |
+
font_demo_images = []
|
126 |
+
|
127 |
+
for i in range(len(font_list)):
|
128 |
+
font_demo_images.append(Image.open(f"demo_fonts/{i}.jpg").convert("RGB"))
|
129 |
+
|
130 |
+
|
131 |
+
def recognize_font(image):
|
132 |
+
transformed_image = transform(image)
|
133 |
+
with torch.no_grad():
|
134 |
+
transformed_image = transformed_image.to(device)
|
135 |
+
output = detector(transformed_image.unsqueeze(0))
|
136 |
+
prob = output[0][: config.FONT_COUNT].softmax(dim=0)
|
137 |
+
|
138 |
+
indicies = torch.topk(prob, 9)[1]
|
139 |
+
|
140 |
+
return [
|
141 |
+
{font_list[i].path: float(prob[i]) for i in range(config.FONT_COUNT)},
|
142 |
+
*[gr.Image.update(value=font_demo_images[indicies[i]]) for i in range(9)],
|
143 |
+
*[
|
144 |
+
gr.Markdown.update(
|
145 |
+
value=f"**Font Name**: {font_list[indicies[i]].path}"
|
146 |
+
)
|
147 |
+
for i in range(9)
|
148 |
+
],
|
149 |
+
]
|
150 |
+
|
151 |
+
|
152 |
+
def generate_grid(num_columns, num_rows):
|
153 |
+
ret_images, ret_labels = [], []
|
154 |
+
with gr.Column():
|
155 |
+
for _ in range(num_rows):
|
156 |
+
with gr.Row():
|
157 |
+
for _ in range(num_columns):
|
158 |
+
with gr.Column():
|
159 |
+
ret_labels.append(gr.Markdown("**Font Name**"))
|
160 |
+
ret_images.append(gr.Image())
|
161 |
+
return ret_images, ret_labels
|
162 |
+
|
163 |
+
|
164 |
+
with gr.Blocks() as demo:
|
165 |
+
with gr.Column():
|
166 |
+
with gr.Row():
|
167 |
+
inp = gr.Image(type="pil", label="Input Image")
|
168 |
+
out = gr.Label(num_top_classes=9, label="Output Font")
|
169 |
+
font_demo_images_blocks, font_demo_labels_blocks = generate_grid(3, 3)
|
170 |
+
|
171 |
+
submit_button = gr.Button(label="Submit")
|
172 |
+
submit_button.click(
|
173 |
+
fn=recognize_font,
|
174 |
+
inputs=inp,
|
175 |
+
outputs=[out, *font_demo_images_blocks, *font_demo_labels_blocks],
|
176 |
+
)
|
177 |
+
|
178 |
+
|
179 |
+
demo.launch(share=args.share)
|