feat: add deepfont baseline

detector/model.py

@@ -10,6 +10,37 @@ import torch.nn as nn
 import pytorch_lightning as ptl
 
 
+class DeepFontBaseline(nn.Module):
+    def __init__(self) -> None:
+        super().__init__()
+        self.model = nn.Sequential(
+            nn.Conv2d(3, 64, 11, 2),
+            nn.BatchNorm2d(64),
+            nn.ReLU(),
+            nn.MaxPool2d(2, 2),
+            nn.Conv2d(64, 128, 3, 1, 1),
+            nn.BatchNorm2d(128),
+            nn.ReLU(),
+            nn.MaxPool2d(2, 2),
+            nn.Conv2d(128, 256, 3, 1, 1),
+            nn.ReLU(),
+            nn.Conv2d(256, 256, 3, 1, 1),
+            nn.ReLU(),
+            nn.Conv2d(256, 256, 3, 1, 1),
+            nn.ReLU(),
+            # fc
+            nn.Flatten(),
+            nn.Linear(256 * 12 * 12, 4096),
+            nn.ReLU(),
+            nn.Linear(4096, 4096),
+            nn.ReLU(),
+            nn.Linear(4096, config.FONT_COUNT),
+        )
+
+    def forward(self, X):
+        return self.model(X)
+
+
 class ResNet18Regressor(nn.Module):
     def __init__(self, pretrained: bool = False, regression_use_tanh: bool = False):
         super().__init__()

train.py

@@ -39,7 +39,7 @@ parser.add_argument(
     "--model",
     type=str,
     default="resnet18",
-    choices=["resnet18", "resnet34", "resnet50", "resnet101"],
+    choices=["resnet18", "resnet34", "resnet50", "resnet101", "deepfont"],
     help="Model to use (default: resnet18)",
 )
 parser.add_argument(
@@ -181,6 +181,11 @@ elif args.model == "resnet101":
     model = ResNet101Regressor(
         pretrained=args.pretrained, 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()