File size: 5,181 Bytes
cd921bb 3d34cc8 cd921bb 474404c cd921bb f27bd8f cd921bb 3d34cc8 | 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 | import csv
from pathlib import Path
import torch
import torch.nn as nn
from datasets import Dataset, Image
from huggingface_hub import PyTorchModelHubMixin
from torch import Tensor
from torch.utils.data import DataLoader
class DetectionHeads(nn.Module):
def __init__(self, input_dim: int, class_num: int):
super().__init__()
self.heads = nn.ModuleList(
[
nn.Sequential(
nn.Linear(input_dim, 64),
nn.GELU(),
nn.Linear(64, 32),
nn.GELU(),
nn.Linear(32, 16),
nn.GELU(),
nn.Linear(16, 8),
nn.GELU(),
)
for _ in range(4)
]
)
self.proj = nn.Linear(8, class_num)
def forward(self, x: Tensor) -> Tensor:
# x: (batch_size, input_dim)
# output: (batch_size, 4, class_num)
y = torch.stack([self.proj(self.heads[i](x)) for i in range(4)], dim=1)
return y
class Baseline2024(nn.Module, PyTorchModelHubMixin):
def __init__(
self,
class_num: int = 26 + 10 + 3,
n_channels: int = 32,
p_dropout: float = 0.95,
):
super().__init__()
self.act = nn.GELU()
self.pool2d = nn.MaxPool2d(kernel_size=2, stride=2, padding=0)
self.pool1d = nn.MaxPool2d(kernel_size=(1, 2), stride=(1, 2), padding=0)
self.conv1 = nn.Conv2d(3, n_channels, kernel_size=3, stride=1, padding=1)
self.bn1 = nn.BatchNorm2d(n_channels)
self.conv2 = nn.Conv2d(
n_channels, n_channels * 2, kernel_size=3, stride=1, padding=1
)
self.bn2 = nn.BatchNorm2d(n_channels * 2)
self.conv3 = nn.Conv2d(
n_channels * 2, n_channels * 4, kernel_size=3, stride=1, padding=1
)
self.bn3 = nn.BatchNorm2d(n_channels * 4)
self.conv4 = nn.Conv2d(
n_channels * 4, n_channels * 8, kernel_size=3, stride=1, padding=1
)
self.bn4 = nn.BatchNorm2d(n_channels * 8)
self.conv5 = nn.Conv2d(
n_channels * 8, n_channels * 16, kernel_size=3, stride=1, padding=1
)
self.bn5 = nn.BatchNorm2d(n_channels * 16)
self.conv6 = nn.Conv2d(
n_channels * 16, n_channels * 32, kernel_size=3, stride=1, padding=1
)
self.bn6 = nn.BatchNorm2d(n_channels * 32)
self.flatten = nn.Flatten()
self.dropout = nn.Dropout(p_dropout)
self.heads = DetectionHeads(n_channels * 32, class_num)
def forward(self, x: Tensor) -> Tensor:
# x: (batch_size, 3, 30, 108)
# output: (batch_size, 4, class_num)
x = self.conv1(x)
x = self.act(x)
x = self.pool2d(x)
x = self.bn1(x)
x = self.conv2(x)
x = self.act(x)
x = self.pool2d(x)
x = self.bn2(x)
x = self.conv3(x)
x = self.act(x)
x = self.pool2d(x)
x = self.bn3(x)
x = self.conv4(x)
x = self.act(x)
x = self.pool2d(x)
x = self.bn4(x)
x = self.conv5(x)
x = self.act(x)
x = self.pool1d(x)
x = self.bn5(x)
x = self.conv6(x)
x = self.act(x)
x = self.pool1d(x)
x = self.bn6(x)
x = self.flatten(x)
x = self.dropout(x)
x = self.heads(x)
return x
char_dict = {
"0": 0,
"1": 1,
"2": 2,
"3": 3,
"4": 4,
"5": 5,
"6": 6,
"7": 7,
"8": 8,
"9": 9,
"-": 10,
"+": 11,
"=": 12,
"a": 13,
"b": 14,
"c": 15,
"d": 16,
"e": 17,
"f": 18,
"g": 19,
"h": 20,
"i": 21,
"j": 22,
"k": 23,
"l": 24,
"m": 25,
"n": 26,
"o": 27,
"p": 28,
"q": 29,
"r": 30,
"s": 31,
"t": 32,
"u": 33,
"v": 34,
"w": 35,
"x": 36,
"y": 37,
"z": 38,
}
char_dict_rev = {v: k for k, v in char_dict.items()}
def tensor_to_text(tensor: torch.Tensor) -> str:
text = ""
for i in tensor:
text += char_dict_rev[torch.argmax(i).item()]
return text
def tensors_to_texts(tensors: torch.Tensor) -> list[str]:
texts = []
for tensor in tensors:
texts.append(tensor_to_text(tensor))
return texts
if __name__ == "__main__":
model = Baseline2024.from_pretrained("./")
dir = Path("/tmp/data/test-data")
captchas = [str(captcha) for captcha in dir.glob("*.jpg")]
dataset = (
Dataset.from_dict({"image": captchas, "path": captchas})
.cast_column("image", Image())
.with_format("torch")
)
loader = DataLoader(dataset, batch_size=16)
model.eval()
submission = "submission.csv"
with open(submission, "w") as f, torch.no_grad():
writer = csv.writer(f)
writer.writerow(["filename", "text"])
for batch in loader:
image = batch["image"].float() / 255.0
output = model(image)
texts = tensors_to_texts(output)
for i, text in enumerate(texts):
file = Path(batch["path"][i]).name
writer.writerow([file, text])
|