File size: 5,237 Bytes
29efa50 c3f3bd0 29efa50 c3f3bd0 29efa50 c3f3bd0 29efa50 |
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 |
import argparse
import os
import gradio as gr
from PIL import Image
from torchvision import transforms
from detector.model import *
from detector import config
from font_dataset.font import load_fonts, load_font_with_exclusion
from huggingface_hub import hf_hub_download
parser = argparse.ArgumentParser()
parser.add_argument(
"-d",
"--device",
type=int,
default=0,
help="GPU devices to use (default: 0), -1 for CPU",
)
parser.add_argument(
"-c",
"--checkpoint",
type=str,
default=None,
help="Trainer checkpoint path (default: None). Use link as huggingface://<user>/<repo>/<file> for huggingface.co models, currently only supports model file in the root directory.",
)
parser.add_argument(
"-m",
"--model",
type=str,
default="resnet18",
choices=["resnet18", "resnet34", "resnet50", "resnet101", "deepfont"],
help="Model to use (default: resnet18)",
)
parser.add_argument(
"-f",
"--font-classification-only",
action="store_true",
help="Font classification only (default: False)",
)
parser.add_argument(
"-z",
"--size",
type=int,
default=512,
help="Model feature image input size (default: 512)",
)
parser.add_argument(
"-s",
"--share",
action="store_true",
help="Get public link via Gradio (default: False)",
)
args = parser.parse_args()
config.INPUT_SIZE = args.size
device = torch.device("cpu") if args.device == -1 else torch.device("cuda", args.device)
regression_use_tanh = False
if args.model == "resnet18":
model = ResNet18Regressor(regression_use_tanh=regression_use_tanh)
elif args.model == "resnet34":
model = ResNet34Regressor(regression_use_tanh=regression_use_tanh)
elif args.model == "resnet50":
model = ResNet50Regressor(regression_use_tanh=regression_use_tanh)
elif args.model == "resnet101":
model = ResNet101Regressor(regression_use_tanh=regression_use_tanh)
elif args.model == "deepfont":
assert args.pretrained is False
assert args.size == 105
assert args.font_classification_only is True
model = DeepFontBaseline()
else:
raise NotImplementedError()
if torch.__version__ >= "2.0" and os.name == "posix":
model = torch.compile(model)
if str(args.checkpoint).startswith("huggingface://"):
args.checkpoint = args.checkpoint[14:]
user, repo, file = args.checkpoint.split("/")
repo = f"{user}/{repo}"
args.checkpoint = hf_hub_download(repo, file)
detector = FontDetector(
model=model,
lambda_font=1,
lambda_direction=1,
lambda_regression=1,
font_classification_only=args.font_classification_only,
lr=1,
betas=(1, 1),
num_warmup_iters=1,
num_iters=1e9,
num_epochs=1e9,
)
detector.load_from_checkpoint(
args.checkpoint,
map_location=device,
model=model,
lambda_font=1,
lambda_direction=1,
lambda_regression=1,
font_classification_only=args.font_classification_only,
lr=1,
betas=(1, 1),
num_warmup_iters=1,
num_iters=1e9,
num_epochs=1e9,
)
detector = detector.to(device)
detector.eval()
transform = transforms.Compose(
[
transforms.Resize((512, 512)),
transforms.ToTensor(),
]
)
print("Preparing fonts ...")
font_list, exclusion_rule = load_fonts()
font_list = list(filter(lambda x: not exclusion_rule(x), font_list))
font_list.sort(key=lambda x: x.path)
for i in range(len(font_list)):
font_list[i].path = font_list[i].path[18:] # remove ./dataset/fonts/./ prefix
font_demo_images = []
for i in range(len(font_list)):
font_demo_images.append(Image.open(f"demo_fonts/{i}.jpg").convert("RGB"))
def recognize_font(image):
transformed_image = transform(image)
with torch.no_grad():
transformed_image = transformed_image.to(device)
output = detector(transformed_image.unsqueeze(0))
prob = output[0][: config.FONT_COUNT].softmax(dim=0)
indicies = torch.topk(prob, 9)[1]
return [
{font_list[i].path: float(prob[i]) for i in range(config.FONT_COUNT)},
*[gr.Image.update(value=font_demo_images[indicies[i]]) for i in range(9)],
*[
gr.Markdown.update(
value=f"**Font Name**: {font_list[indicies[i]].path}"
)
for i in range(9)
],
]
def generate_grid(num_columns, num_rows):
ret_images, ret_labels = [], []
with gr.Column():
for _ in range(num_rows):
with gr.Row():
for _ in range(num_columns):
with gr.Column():
ret_labels.append(gr.Markdown("**Font Name**"))
ret_images.append(gr.Image())
return ret_images, ret_labels
with gr.Blocks() as demo:
with gr.Column():
with gr.Row():
inp = gr.Image(type="pil", label="Input Image")
out = gr.Label(num_top_classes=9, label="Output Font")
font_demo_images_blocks, font_demo_labels_blocks = generate_grid(3, 3)
submit_button = gr.Button(label="Submit")
submit_button.click(
fn=recognize_font,
inputs=inp,
outputs=[out, *font_demo_images_blocks, *font_demo_labels_blocks],
)
demo.launch(share=args.share)
|