Upload 149 files

milk10k_effb2_metadata/MILK10K_EFFB2_METADATA_CLI.md

@@ -50,6 +50,21 @@ python train_milk10k_effb2_dual_metadata.py \
   --output-dir milk10k_effb2_baseline
 ```
 
+## ConvNeXt Base
+
+Use the dedicated `DualConvNeXtMetadataClassifier` with two ImageNet-initialized
+ConvNeXt Base encoders. When `--image-size` is omitted, ConvNeXt uses 384x384.
+
+```bash
+python train_milk10k_effb2_dual_metadata.py \
+  --backbone convnext_base \
+  --batch-size 4 \
+  --amp \
+  --output-dir milk10k_convnext_base_metadata
+```
+
+Pass `--image-size` explicitly to override the 384x384 default.
+
 ## Metadata Fusion Options
 
 Keep the baseline concat fusion:

milk10k_effb2_metadata/cli.py

@@ -1,4 +1,4 @@
-"""CLI for the EfficientNet-B2 dual metadata trainer."""
+"""CLI for the dual-backbone metadata trainer."""
 
 from __future__ import annotations
 
@@ -7,7 +7,7 @@ from pathlib import Path
 
 
 def parse_args() -> argparse.Namespace:
-    parser = argparse.ArgumentParser(description="Train MILK10k dual EfficientNet-B2 with metadata fusion.")
+    parser = argparse.ArgumentParser(description="Train MILK10k dual-image backbones with metadata fusion.")
     parser.add_argument("--data-dir", type=Path, default=None)
     parser.add_argument(
         "--clinical-checkpoint",
@@ -25,7 +25,7 @@ def parse_args() -> argparse.Namespace:
         "--resume-checkpoint",
         type=Path,
         default=None,
-        help="Resume model weights/best score from an EffB2 metadata checkpoint, usually output-dir/best.pt.",
+        help="Resume model weights/best score from a metadata checkpoint, usually output-dir/best.pt.",
     )
     parser.add_argument("--output-dir", type=Path, default=Path("milk10k_dual_effb2_metadata_runs"))
     parser.add_argument("--freeze-epochs", type=int, default=8)

milk10k_effb2_metadata/engine.py

@@ -166,6 +166,7 @@ def save_checkpoint(
         "epoch": epoch,
         "phase": phase,
         "model_state": model.state_dict(),
+        "model_type": model.__class__.__name__,
         "optimizer_state": optimizer.state_dict(),
         "best_val_f1_macro": best_val_f1,
         "best_selection_metric": best_val_f1,

milk10k_effb2_metadata/inference.py

@@ -1,4 +1,4 @@
-"""Inference CLI for EffB2 dual metadata checkpoints."""
+"""Inference CLI for dual-image metadata checkpoints."""
 
 from __future__ import annotations
 
@@ -17,7 +17,12 @@ from tqdm.auto import tqdm
 from datasets import LABEL_COLUMNS, normalize_image_type
 from milk10k_effb2_metadata.data import METADATA_COLUMNS, make_transforms, metadata_vector, resolve_monet_columns
 from milk10k_effb2_metadata.metrics import apply_class_bias, compute_metrics
-from milk10k_effb2_metadata.models import DualEffB2MetadataClassifier
+from milk10k_effb2_metadata.models import (
+    DualEffB2MetadataClassifier,
+    model_class_for_backbone,
+    normalize_backbone_name,
+    resolve_image_size,
+)
 from milk10k_effb2_metadata.training import json_safe
 
 
@@ -47,7 +52,7 @@ class InferencePairedDataset(Dataset):
 
 
 def parse_args() -> argparse.Namespace:
-    parser = argparse.ArgumentParser(description="Run inference with a MILK10k dual EffB2 metadata checkpoint.")
+    parser = argparse.ArgumentParser(description="Run inference with a MILK10k dual-image metadata checkpoint.")
     parser.add_argument("--checkpoint", type=Path, nargs="*", default=None, help="One or more checkpoint paths.")
     parser.add_argument(
         "--checkpoint-dir",
@@ -170,7 +175,15 @@ def build_model_from_checkpoint(checkpoint: dict[str, Any], metadata_dim: int, d
     class_names = checkpoint["class_names"]
     clinical_backend = infer_backend_from_model_state(state, "clinical_encoder.")
     dermoscopic_backend = infer_backend_from_model_state(state, "dermoscopic_encoder.")
-    model = DualEffB2MetadataClassifier(
+    backbone = normalize_backbone_name(checkpoint_arg(checkpoint_args, "backbone", "efficientnet_b2"))
+    model_class = model_class_for_backbone(backbone)
+    saved_model_type = checkpoint.get("model_type")
+    if saved_model_type is not None and saved_model_type != model_class.__name__:
+        raise ValueError(
+            f"Checkpoint model_type {saved_model_type!r} does not match backbone "
+            f"{backbone!r} ({model_class.__name__})."
+        )
+    model = model_class(
         num_classes=len(class_names),
         metadata_input_dim=metadata_dim,
         branch_dim=checkpoint_arg(checkpoint_args, "branch_dim", 512),
@@ -180,7 +193,7 @@ def build_model_from_checkpoint(checkpoint: dict[str, Any], metadata_dim: int, d
         imagenet_pretrained=False,
         clinical_backbone_backend=clinical_backend,
         dermoscopic_backbone_backend=dermoscopic_backend,
-        backbone=checkpoint_arg(checkpoint_args, "backbone", "efficientnet_b2"),
+        backbone=backbone,
         disable_metadata=checkpoint_arg(checkpoint_args, "disable_metadata", False),
         metadata_fusion=checkpoint_arg(checkpoint_args, "metadata_fusion", "concat"),
         image_fusion=checkpoint_arg(checkpoint_args, "image_fusion", "concat"),
@@ -295,7 +308,12 @@ def main() -> None:
                 f"Checkpoint class_names mismatch: expected {class_names}, got {checkpoint_class_names} from {checkpoint_path}"
             )
         checkpoint_args = checkpoint.get("args", {})
-        image_size = args.image_size or int(checkpoint_args.get("image_size", 260))
+        backbone = checkpoint_args.get("backbone", "efficientnet_b2")
+        checkpoint_image_size = checkpoint_args.get("image_size")
+        image_size = resolve_image_size(
+            backbone,
+            args.image_size if args.image_size is not None else checkpoint_image_size,
+        )
         _, eval_transform = make_transforms(image_size)
         dataset = InferencePairedDataset(df, checkpoint["metadata_spec"], eval_transform)
         loader = DataLoader(

milk10k_effb2_metadata/milk10k_effb2_metadata/MILK10K_EFFB2_METADATA_CLI.md

@@ -50,6 +50,21 @@ python train_milk10k_effb2_dual_metadata.py \
   --output-dir milk10k_effb2_baseline
 ```
 
+## ConvNeXt Base
+
+Use the dedicated `DualConvNeXtMetadataClassifier` with two ImageNet-initialized
+ConvNeXt Base encoders. When `--image-size` is omitted, ConvNeXt uses 384x384.
+
+```bash
+python train_milk10k_effb2_dual_metadata.py \
+  --backbone convnext_base \
+  --batch-size 4 \
+  --amp \
+  --output-dir milk10k_convnext_base_metadata
+```
+
+Pass `--image-size` explicitly to override the 384x384 default.
+
 ## Metadata Fusion Options
 
 Keep the baseline concat fusion:

milk10k_effb2_metadata/milk10k_effb2_metadata/cli.py

@@ -1,4 +1,4 @@
-"""CLI for the EfficientNet-B2 dual metadata trainer."""
+"""CLI for the dual-backbone metadata trainer."""
 
 from __future__ import annotations
 
@@ -7,7 +7,7 @@ from pathlib import Path
 
 
 def parse_args() -> argparse.Namespace:
-    parser = argparse.ArgumentParser(description="Train MILK10k dual EfficientNet-B2 with metadata fusion.")
+    parser = argparse.ArgumentParser(description="Train MILK10k dual-image backbones with metadata fusion.")
     parser.add_argument("--data-dir", type=Path, default=None)
     parser.add_argument(
         "--clinical-checkpoint",
@@ -25,7 +25,7 @@ def parse_args() -> argparse.Namespace:
         "--resume-checkpoint",
         type=Path,
         default=None,
-        help="Resume model weights/best score from an EffB2 metadata checkpoint, usually output-dir/best.pt.",
+        help="Resume model weights/best score from a metadata checkpoint, usually output-dir/best.pt.",
     )
     parser.add_argument("--output-dir", type=Path, default=Path("milk10k_dual_effb2_metadata_runs"))
     parser.add_argument("--freeze-epochs", type=int, default=8)

milk10k_effb2_metadata/milk10k_effb2_metadata/engine.py

@@ -166,6 +166,7 @@ def save_checkpoint(
         "epoch": epoch,
         "phase": phase,
         "model_state": model.state_dict(),
+        "model_type": model.__class__.__name__,
         "optimizer_state": optimizer.state_dict(),
         "best_val_f1_macro": best_val_f1,
         "best_selection_metric": best_val_f1,

milk10k_effb2_metadata/milk10k_effb2_metadata/inference.py

@@ -1,4 +1,4 @@
-"""Inference CLI for EffB2 dual metadata checkpoints."""
+"""Inference CLI for dual-image metadata checkpoints."""
 
 from __future__ import annotations
 
@@ -17,7 +17,12 @@ from tqdm.auto import tqdm
 from datasets import LABEL_COLUMNS, normalize_image_type
 from milk10k_effb2_metadata.data import METADATA_COLUMNS, make_transforms, metadata_vector, resolve_monet_columns
 from milk10k_effb2_metadata.metrics import apply_class_bias, compute_metrics
-from milk10k_effb2_metadata.models import DualEffB2MetadataClassifier
+from milk10k_effb2_metadata.models import (
+    DualEffB2MetadataClassifier,
+    model_class_for_backbone,
+    normalize_backbone_name,
+    resolve_image_size,
+)
 from milk10k_effb2_metadata.training import json_safe
 
 
@@ -47,7 +52,7 @@ class InferencePairedDataset(Dataset):
 
 
 def parse_args() -> argparse.Namespace:
-    parser = argparse.ArgumentParser(description="Run inference with a MILK10k dual EffB2 metadata checkpoint.")
+    parser = argparse.ArgumentParser(description="Run inference with a MILK10k dual-image metadata checkpoint.")
     parser.add_argument("--checkpoint", type=Path, nargs="*", default=None, help="One or more checkpoint paths.")
     parser.add_argument(
         "--checkpoint-dir",
@@ -170,7 +175,15 @@ def build_model_from_checkpoint(checkpoint: dict[str, Any], metadata_dim: int, d
     class_names = checkpoint["class_names"]
     clinical_backend = infer_backend_from_model_state(state, "clinical_encoder.")
     dermoscopic_backend = infer_backend_from_model_state(state, "dermoscopic_encoder.")
-    model = DualEffB2MetadataClassifier(
+    backbone = normalize_backbone_name(checkpoint_arg(checkpoint_args, "backbone", "efficientnet_b2"))
+    model_class = model_class_for_backbone(backbone)
+    saved_model_type = checkpoint.get("model_type")
+    if saved_model_type is not None and saved_model_type != model_class.__name__:
+        raise ValueError(
+            f"Checkpoint model_type {saved_model_type!r} does not match backbone "
+            f"{backbone!r} ({model_class.__name__})."
+        )
+    model = model_class(
         num_classes=len(class_names),
         metadata_input_dim=metadata_dim,
         branch_dim=checkpoint_arg(checkpoint_args, "branch_dim", 512),
@@ -180,7 +193,7 @@ def build_model_from_checkpoint(checkpoint: dict[str, Any], metadata_dim: int, d
         imagenet_pretrained=False,
         clinical_backbone_backend=clinical_backend,
         dermoscopic_backbone_backend=dermoscopic_backend,
-        backbone=checkpoint_arg(checkpoint_args, "backbone", "efficientnet_b2"),
+        backbone=backbone,
         disable_metadata=checkpoint_arg(checkpoint_args, "disable_metadata", False),
         metadata_fusion=checkpoint_arg(checkpoint_args, "metadata_fusion", "concat"),
         image_fusion=checkpoint_arg(checkpoint_args, "image_fusion", "concat"),
@@ -295,7 +308,12 @@ def main() -> None:
                 f"Checkpoint class_names mismatch: expected {class_names}, got {checkpoint_class_names} from {checkpoint_path}"
             )
         checkpoint_args = checkpoint.get("args", {})
-        image_size = args.image_size or int(checkpoint_args.get("image_size", 260))
+        backbone = checkpoint_args.get("backbone", "efficientnet_b2")
+        checkpoint_image_size = checkpoint_args.get("image_size")
+        image_size = resolve_image_size(
+            backbone,
+            args.image_size if args.image_size is not None else checkpoint_image_size,
+        )
         _, eval_transform = make_transforms(image_size)
         dataset = InferencePairedDataset(df, checkpoint["metadata_spec"], eval_transform)
         loader = DataLoader(

milk10k_effb2_metadata/milk10k_effb2_metadata/model_setup.py

@@ -12,7 +12,7 @@ from milk10k_effb2_metadata.checkpoints import (
     load_encoder_checkpoint,
     resolve_backbone_backends,
 )
-from milk10k_effb2_metadata.models import DualEffB2MetadataClassifier
+from milk10k_effb2_metadata.models import DualEffB2MetadataClassifier, model_class_for_backbone
 
 
 def infer_branch_backend_from_state(state: dict[str, torch.Tensor], branch_prefix: str) -> str:
@@ -136,7 +136,8 @@ def build_model(
     clinical_backbone_backend: str,
     dermoscopic_backbone_backend: str,
 ) -> DualEffB2MetadataClassifier:
-    model = DualEffB2MetadataClassifier(
+    model_class = model_class_for_backbone(args.backbone)
+    model = model_class(
         num_classes=len(class_names),
         metadata_input_dim=metadata_dim,
         branch_dim=args.branch_dim,

milk10k_effb2_metadata/milk10k_effb2_metadata/models.py

@@ -386,6 +386,19 @@ class DualEffB2MetadataClassifier(nn.Module):
         return F.adaptive_avg_pool2d(gated, 1)
 
 
+class DualConvNeXtMetadataClassifier(DualEffB2MetadataClassifier):
+    """Dual-image metadata classifier backed by independent ConvNeXt Base encoders."""
+
+    def __init__(self, *args, **kwargs) -> None:
+        backbone = normalize_backbone_name(kwargs.pop("backbone", "convnext_base"))
+        if backbone != "convnext_base":
+            raise ValueError(
+                "DualConvNeXtMetadataClassifier only supports the convnext_base backbone, "
+                f"got {backbone!r}."
+            )
+        super().__init__(*args, backbone=backbone, **kwargs)
+
+
 def normalize_backbone_name(name: str) -> str:
     name = name.lower().replace(" ", "").replace("_", "").replace("-", "")
     if name in ("efficientnetb2", "effnetb2", "effb2"):
@@ -399,6 +412,31 @@ def normalize_backbone_name(name: str) -> str:
     raise ValueError(f"Unknown backbone: {name}")
 
 
+def model_class_for_backbone(backbone: str) -> type[DualEffB2MetadataClassifier]:
+    """Return the dedicated model class for a normalized backbone name."""
+    backbone = normalize_backbone_name(backbone)
+    if backbone == "convnext_base":
+        return DualConvNeXtMetadataClassifier
+    return DualEffB2MetadataClassifier
+
+
+def default_image_size(backbone: str) -> int:
+    """Return the training/inference resolution used when --image-size is omitted."""
+    backbone = normalize_backbone_name(backbone)
+    if backbone == "efficientnet_b2":
+        return 260
+    if backbone == "efficientnet_b1":
+        return 240
+    if backbone == "convnext_base":
+        return 384
+    return 224
+
+
+def resolve_image_size(backbone: str, image_size: int | None) -> int:
+    """Use an explicit image size when provided, otherwise use the backbone default."""
+    return int(image_size) if image_size is not None else default_image_size(backbone)
+
+
 def extract_spatial_features(encoder: nn.Module, backbone_backend: str, backbone: str, images: torch.Tensor) -> torch.Tensor:
     if backbone_backend == "timm":
         features = encoder.forward_features(images)

milk10k_effb2_metadata/milk10k_effb2_metadata/training.py

@@ -13,7 +13,7 @@ def run(args: argparse.Namespace) -> None:
     from datasets import resolve_data_dir, set_seed
     from milk10k_effb2_metadata.data import load_paired_dataframe
     from milk10k_effb2_metadata.model_setup import resolve_training_backbone_backends
-    from milk10k_effb2_metadata.models import normalize_backbone_name
+    from milk10k_effb2_metadata.models import normalize_backbone_name, resolve_image_size
     from milk10k_effb2_metadata.runner import train_kfold, train_single_run
 
     if args.k_folds < 1:
@@ -28,13 +28,7 @@ def run(args: argparse.Namespace) -> None:
         args.metadata_gate_hidden_dim = args.metadata_dim
     if args.resume_checkpoint is None and args.clinical_checkpoint is None and args.dermoscopic_checkpoint is None:
         args.imagenet_pretrained = True
-    if args.image_size is None:
-        if args.backbone == "efficientnet_b2":
-            args.image_size = 260
-        elif args.backbone == "efficientnet_b1":
-            args.image_size = 240
-        else:  # resnet50, convnext_base
-            args.image_size = 224
+    args.image_size = resolve_image_size(args.backbone, args.image_size)
 
     df = load_paired_dataframe(data_dir)
     class_names = sorted(df["label"].unique())

milk10k_effb2_metadata/milk10k_effb2_metadata/training_utils.py

@@ -30,6 +30,7 @@ def save_run_config(
         "args": json_safe(vars(args)),
         "class_names": class_names,
         "metadata_spec": json_safe(metadata_spec),
+        "model_type": "DualConvNeXtMetadataClassifier" if args.backbone == "convnext_base" else "DualEffB2MetadataClassifier",
         "train_size": len(train_df),
         "val_size": len(val_df),
         "fold": fold,

milk10k_effb2_metadata/model_setup.py

@@ -12,7 +12,7 @@ from milk10k_effb2_metadata.checkpoints import (
     load_encoder_checkpoint,
     resolve_backbone_backends,
 )
-from milk10k_effb2_metadata.models import DualEffB2MetadataClassifier
+from milk10k_effb2_metadata.models import DualEffB2MetadataClassifier, model_class_for_backbone
 
 
 def infer_branch_backend_from_state(state: dict[str, torch.Tensor], branch_prefix: str) -> str:
@@ -136,7 +136,8 @@ def build_model(
     clinical_backbone_backend: str,
     dermoscopic_backbone_backend: str,
 ) -> DualEffB2MetadataClassifier:
-    model = DualEffB2MetadataClassifier(
+    model_class = model_class_for_backbone(args.backbone)
+    model = model_class(
         num_classes=len(class_names),
         metadata_input_dim=metadata_dim,
         branch_dim=args.branch_dim,

milk10k_effb2_metadata/models.py

@@ -386,6 +386,19 @@ class DualEffB2MetadataClassifier(nn.Module):
         return F.adaptive_avg_pool2d(gated, 1)
 
 
+class DualConvNeXtMetadataClassifier(DualEffB2MetadataClassifier):
+    """Dual-image metadata classifier backed by independent ConvNeXt Base encoders."""
+
+    def __init__(self, *args, **kwargs) -> None:
+        backbone = normalize_backbone_name(kwargs.pop("backbone", "convnext_base"))
+        if backbone != "convnext_base":
+            raise ValueError(
+                "DualConvNeXtMetadataClassifier only supports the convnext_base backbone, "
+                f"got {backbone!r}."
+            )
+        super().__init__(*args, backbone=backbone, **kwargs)
+
+
 def normalize_backbone_name(name: str) -> str:
     name = name.lower().replace(" ", "").replace("_", "").replace("-", "")
     if name in ("efficientnetb2", "effnetb2", "effb2"):
@@ -399,6 +412,31 @@ def normalize_backbone_name(name: str) -> str:
     raise ValueError(f"Unknown backbone: {name}")
 
 
+def model_class_for_backbone(backbone: str) -> type[DualEffB2MetadataClassifier]:
+    """Return the dedicated model class for a normalized backbone name."""
+    backbone = normalize_backbone_name(backbone)
+    if backbone == "convnext_base":
+        return DualConvNeXtMetadataClassifier
+    return DualEffB2MetadataClassifier
+
+
+def default_image_size(backbone: str) -> int:
+    """Return the training/inference resolution used when --image-size is omitted."""
+    backbone = normalize_backbone_name(backbone)
+    if backbone == "efficientnet_b2":
+        return 260
+    if backbone == "efficientnet_b1":
+        return 240
+    if backbone == "convnext_base":
+        return 384
+    return 224
+
+
+def resolve_image_size(backbone: str, image_size: int | None) -> int:
+    """Use an explicit image size when provided, otherwise use the backbone default."""
+    return int(image_size) if image_size is not None else default_image_size(backbone)
+
+
 def extract_spatial_features(encoder: nn.Module, backbone_backend: str, backbone: str, images: torch.Tensor) -> torch.Tensor:
     if backbone_backend == "timm":
         features = encoder.forward_features(images)

milk10k_effb2_metadata/tests/test_fusion_and_f1_loss.py

@@ -21,7 +21,14 @@ if MISSING_DEPENDENCY is None:
         from PIL import Image
         from milk10k_effb2_metadata.data import PairedMilk10kMetadataDataset
         from milk10k_effb2_metadata.losses import SoftMacroF1Loss, f1_class_weight_tensor
-        from milk10k_effb2_metadata.models import DualEffB2MetadataClassifier
+        from milk10k_effb2_metadata.inference import build_model_from_checkpoint
+        from milk10k_effb2_metadata.models import (
+            DualConvNeXtMetadataClassifier,
+            DualEffB2MetadataClassifier,
+            default_image_size,
+            model_class_for_backbone,
+            resolve_image_size,
+        )
     except ModuleNotFoundError as exc:  # pragma: no cover - local minimal env may omit ML deps.
         MISSING_DEPENDENCY = exc.name
 
@@ -54,6 +61,76 @@ if MISSING_DEPENDENCY is None:
 
 
     class FusionSmokeTest(unittest.TestCase):
+        @staticmethod
+        def model_kwargs(backbone: str = "efficientnet_b2") -> dict:
+            return {
+                "num_classes": 4,
+                "metadata_input_dim": 5,
+                "branch_dim": 8,
+                "metadata_dim": 6,
+                "classifier_hidden_dim": 12,
+                "dropout": 0.0,
+                "imagenet_pretrained": False,
+                "clinical_backbone_backend": "timm",
+                "dermoscopic_backbone_backend": "timm",
+                "backbone": backbone,
+            }
+
+        def test_dedicated_convnext_forward_and_selection(self) -> None:
+            with patch("milk10k_effb2_metadata.models.build_feature_encoder", side_effect=fake_build_feature_encoder):
+                model = DualConvNeXtMetadataClassifier(**self.model_kwargs("convnext_base"))
+                logits = model(
+                    torch.randn(2, 3, 8, 8),
+                    torch.randn(2, 3, 8, 8),
+                    torch.randn(2, 5),
+                )
+            self.assertEqual(tuple(logits.shape), (2, 4))
+            self.assertIs(model_class_for_backbone("convnext_base"), DualConvNeXtMetadataClassifier)
+            self.assertIs(model_class_for_backbone("efficientnet_b2"), DualEffB2MetadataClassifier)
+
+        def test_backbone_default_image_sizes(self) -> None:
+            self.assertEqual(default_image_size("convnext_base"), 384)
+            self.assertEqual(default_image_size("efficientnet_b2"), 260)
+            self.assertEqual(default_image_size("efficientnet_b1"), 240)
+            self.assertEqual(default_image_size("resnet50"), 224)
+            self.assertEqual(resolve_image_size("convnext_base", 320), 320)
+
+        def test_checkpoint_reconstructs_dedicated_and_legacy_models(self) -> None:
+            checkpoint_args = {
+                "branch_dim": 8,
+                "metadata_dim": 6,
+                "classifier_hidden_dim": 12,
+                "dropout": 0.0,
+                "metadata_fusion": "concat",
+                "image_fusion": "concat",
+                "logit_fusion_mode": "single",
+            }
+            with patch("milk10k_effb2_metadata.models.build_feature_encoder", side_effect=fake_build_feature_encoder):
+                convnext = DualConvNeXtMetadataClassifier(**self.model_kwargs("convnext_base"))
+                efficientnet = DualEffB2MetadataClassifier(**self.model_kwargs("efficientnet_b2"))
+                with patch("milk10k_effb2_metadata.inference.infer_backend_from_model_state", return_value="timm"):
+                    loaded_convnext = build_model_from_checkpoint(
+                        {
+                            "model_state": convnext.state_dict(),
+                            "model_type": "DualConvNeXtMetadataClassifier",
+                            "class_names": ["A", "B", "C", "D"],
+                            "args": {**checkpoint_args, "backbone": "convnext_base"},
+                        },
+                        metadata_dim=5,
+                        device=torch.device("cpu"),
+                    )
+                    loaded_legacy = build_model_from_checkpoint(
+                        {
+                            "model_state": efficientnet.state_dict(),
+                            "class_names": ["A", "B", "C", "D"],
+                            "args": {**checkpoint_args, "backbone": "efficientnet_b2"},
+                        },
+                        metadata_dim=5,
+                        device=torch.device("cpu"),
+                    )
+            self.assertIsInstance(loaded_convnext, DualConvNeXtMetadataClassifier)
+            self.assertIs(type(loaded_legacy), DualEffB2MetadataClassifier)
+
         def test_all_image_and_metadata_fusions_forward(self) -> None:
             modes = [
                 "concat",

milk10k_effb2_metadata/training.py

@@ -13,7 +13,7 @@ def run(args: argparse.Namespace) -> None:
     from datasets import resolve_data_dir, set_seed
     from milk10k_effb2_metadata.data import load_paired_dataframe
     from milk10k_effb2_metadata.model_setup import resolve_training_backbone_backends
-    from milk10k_effb2_metadata.models import normalize_backbone_name
+    from milk10k_effb2_metadata.models import normalize_backbone_name, resolve_image_size
     from milk10k_effb2_metadata.runner import train_kfold, train_single_run
 
     if args.k_folds < 1:
@@ -28,13 +28,7 @@ def run(args: argparse.Namespace) -> None:
         args.metadata_gate_hidden_dim = args.metadata_dim
     if args.resume_checkpoint is None and args.clinical_checkpoint is None and args.dermoscopic_checkpoint is None:
         args.imagenet_pretrained = True
-    if args.image_size is None:
-        if args.backbone == "efficientnet_b2":
-            args.image_size = 260
-        elif args.backbone == "efficientnet_b1":
-            args.image_size = 240
-        else:  # resnet50, convnext_base
-            args.image_size = 224
+    args.image_size = resolve_image_size(args.backbone, args.image_size)
 
     df = load_paired_dataframe(data_dir)
     class_names = sorted(df["label"].unique())

milk10k_effb2_metadata/training_utils.py

@@ -35,6 +35,7 @@ def save_run_config(
         "class_names": class_names,
         "label_to_idx": label_to_idx,
         "metadata_spec": json_safe(metadata_spec),
+        "model_type": "DualConvNeXtMetadataClassifier" if args.backbone == "convnext_base" else "DualEffB2MetadataClassifier",
         "train_size": len(train_df),
         "val_size": len(val_df),
         "fold": fold,