Fix/lint errors

detectzoo/__init__.py

@@ -1,49 +1,25 @@
 """DetectZoo: A unified toolkit for detecting AI-generated content."""
 
-import importlib
-import warnings
-
-from detectzoo.utils.hf_quiet import configure_hf_quiet
-
-configure_hf_quiet()
-
-
-def _try_import(module: str) -> None:
-    """Import a modality subpackage, warning (not failing) if optional deps are missing.
-
-    This keeps pure-audio or pure-text workflows usable even when optional
-    image/text/audio extras are not installed.
-    """
-    try:
-        importlib.import_module(module)
-    except ImportError as exc:
-        warnings.warn(
-            f"detectzoo: skipped loading '{module}' ({exc}). "
-            "Install the corresponding optional extra to enable it "
-            "(e.g. `pip install detectzoo[audio]`).",
-            stacklevel=2,
-        )
-
-
-for _mod in (
-    "detectzoo.datasets.audio",
-    "detectzoo.datasets.image",
-    "detectzoo.datasets.text",
-    "detectzoo.detectors.audio",
-    "detectzoo.detectors.image",
-    "detectzoo.detectors.text",
-):
-    _try_import(_mod)
-
-from detectzoo.core.base import BaseDetector, DetectionResult  # noqa: E402
-from detectzoo.core.registry import (  # noqa: E402
+# isort: off
+import detectzoo.utils.hf_quiet  # noqa: F401
+# isort: on
+
+# Eager-load modality subpackages so @register_detector / @register_dataset run.
+import detectzoo.datasets.audio  # noqa: F401
+import detectzoo.datasets.image  # noqa: F401
+import detectzoo.datasets.text  # noqa: F401
+import detectzoo.detectors.audio  # noqa: F401
+import detectzoo.detectors.image  # noqa: F401
+import detectzoo.detectors.text  # noqa: F401
+from detectzoo.core.base import BaseDetector, DetectionResult
+from detectzoo.core.registry import (
     list_datasets,
     list_detectors,
     load_dataset,
     load_detector,
 )
 
-__version__ = "0.1.0"
+__version__ = "0.1.4"
 
 __all__ = [
     "BaseDetector",

detectzoo/benchmarks/evaluator.py

@@ -22,8 +22,15 @@
 # anti-spoofing trio (EER / AUC / F1); image and text keep DetectZoo's
 # original column set unchanged.
 _DEFAULT_PRINT_COLUMNS = [
-    "detector", "accuracy", "precision", "recall", "f1",
-    "tpr", "fpr", "roc_auc", "pr_auc",
+    "detector",
+    "accuracy",
+    "precision",
+    "recall",
+    "f1",
+    "tpr",
+    "fpr",
+    "roc_auc",
+    "pr_auc",
 ]
 _PRINT_VIEWS = {
     "audio": ["detector", "eer", "roc_auc", "f1"],
@@ -98,10 +105,7 @@ def evaluate_single(
         metrics["n_samples"] = len(labels)
 
         if save_scores:
-            metrics["samples"] = [
-                {"label": lbl, "score": scr}
-                for lbl, scr in zip(labels, scores)
-            ]
+            metrics["samples"] = [{"label": lbl, "score": scr} for lbl, scr in zip(labels, scores)]
 
         return metrics
 
@@ -159,8 +163,12 @@ def run_and_print(self, detectors: Sequence[BaseDetector]) -> None:
         print(header)
         print("-" * len(header))
         for metrics in all_results.values():
-            row = " | ".join(f"{metrics.get(k, ''):>18}" if isinstance(metrics.get(k), str)
-                             else f"{metrics.get(k, 0):>18.4f}" for k in header_keys)
+            row = " | ".join(
+                f"{metrics.get(k, ''):>18}"
+                if isinstance(metrics.get(k), str)
+                else f"{metrics.get(k, 0):>18.4f}"
+                for k in header_keys
+            )
             print(row)
 
     def _save_payload(

detectzoo/core/registry.py

@@ -56,19 +56,15 @@ def load_detector(name: str, **kwargs: Any) -> BaseDetector:
     resolved = _ALIASES.get(name, name)
     if resolved not in _REGISTRY:
         available = ", ".join(sorted(set(_REGISTRY) | set(_ALIASES))) or "(none)"
-        raise ValueError(
-            f"Unknown detector '{name}'. Available detectors: {available}"
-        )
+        raise ValueError(f"Unknown detector '{name}'. Available detectors: {available}")
     return _REGISTRY[resolved](**kwargs)
 
 
 def list_detectors(modality: str | None = None) -> list[str]:
     """Return names of all registered detectors, optionally filtered by modality."""
     if modality is None:
         return sorted(_REGISTRY)
-    return sorted(
-        name for name, cls in _REGISTRY.items() if cls.modality == modality
-    )
+    return sorted(name for name, cls in _REGISTRY.items() if cls.modality == modality)
 
 
 # ======================================================================
@@ -127,16 +123,12 @@ def load_dataset(name: str, **kwargs: Any) -> BaseDataset:
     resolved = _DATASET_ALIASES.get(name, name)
     if resolved not in _DATASET_REGISTRY:
         available = ", ".join(sorted(set(_DATASET_REGISTRY) | set(_DATASET_ALIASES))) or "(none)"
-        raise ValueError(
-            f"Unknown dataset '{name}'. Available datasets: {available}"
-        )
+        raise ValueError(f"Unknown dataset '{name}'. Available datasets: {available}")
     return _DATASET_REGISTRY[resolved](**kwargs)
 
 
 def list_datasets(modality: str | None = None) -> list[str]:
     """Return names of all registered datasets, optionally filtered by modality."""
     if modality is None:
         return sorted(_DATASET_REGISTRY)
-    return sorted(
-        name for name, cls in _DATASET_REGISTRY.items() if cls.modality == modality
-    )
+    return sorted(name for name, cls in _DATASET_REGISTRY.items() if cls.modality == modality)

detectzoo/datasets/__init__.py

@@ -52,4 +52,3 @@
     "WritingPromptsDataset",
     "XSumDataset",
 ]
-

detectzoo/datasets/_download.py

@@ -100,9 +100,9 @@ def download_and_extract_tar(
     *,
     force: bool = False,
 ) -> Path:
-    """Download a tar archive (optionally gzip/bzip2/xz-compressed), extract it, and cache the result.
+    """Download a tar archive (gzip/bzip2/xz optional), extract it, and cache the result.
 
-    A ``.download_complete`` marker file prevents re-downloading on 
+    A ``.download_complete`` marker file prevents re-downloading on
     subsequent calls.
     """
     marker = dest_dir / ".download_complete"

detectzoo/datasets/audio/asvspoof2019.py

@@ -89,10 +89,7 @@ def _looks_like_track_root(d: Path) -> bool:
         if (d / proto_leaf).is_dir():
             return True
         # Tolerate extracts that contain only flac dirs (no protocols).
-        return any(
-            (d / leaf / "flac").is_dir()
-            for leaf in (train_leaf, dev_leaf, eval_leaf)
-        )
+        return any((d / leaf / "flac").is_dir() for leaf in (train_leaf, dev_leaf, eval_leaf))
 
     # User pointed straight at one of the partition leaves.
     if u.name in {train_leaf, dev_leaf, eval_leaf}:
@@ -119,9 +116,7 @@ def _looks_like_track_root(d: Path) -> bool:
 def _protocol_file(track_root: Path, track: str, partition: str) -> Path:
     proto_dir = track_root / f"ASVspoof2019_{track}_cm_protocols"
     if not proto_dir.is_dir():
-        raise FileNotFoundError(
-            f"ASVspoof 2019: missing protocols directory {proto_dir}"
-        )
+        raise FileNotFoundError(f"ASVspoof 2019: missing protocols directory {proto_dir}")
     if partition == "train":
         names = (f"ASVspoof2019.{track}.cm.train.trn.txt",)
     elif partition == "dev":
@@ -143,9 +138,7 @@ def _flac_dir(track_root: Path, track: str, partition: str) -> Path:
     sub = "train" if partition == "train" else "dev" if partition == "dev" else "eval"
     d = track_root / f"ASVspoof2019_{track}_{sub}" / "flac"
     if not d.is_dir():
-        raise FileNotFoundError(
-            f"ASVspoof 2019: expected FLAC directory {d}"
-        )
+        raise FileNotFoundError(f"ASVspoof 2019: expected FLAC directory {d}")
     return d
 
 

detectzoo/datasets/audio/deepfake_eval_2024.py

@@ -127,9 +127,7 @@ def _load_from_metadata(
     meta_path = _find_metadata_csv(root)
     audio_dir = root / _AUDIO_SUBDIR
     if not audio_dir.is_dir():
-        raise FileNotFoundError(
-            f"Deepfake-Eval-2024: expected audio directory {audio_dir}"
-        )
+        raise FileNotFoundError(f"Deepfake-Eval-2024: expected audio directory {audio_dir}")
 
     items: List[DatasetItem] = []
     missing: List[str] = []
@@ -161,8 +159,7 @@ def _load_from_metadata(
                 if skip_missing:
                     continue
                 raise FileNotFoundError(
-                    f"Deepfake-Eval-2024: audio missing for {filename!r} "
-                    f"(looked under {audio_dir})"
+                    f"Deepfake-Eval-2024: audio missing for {filename!r} (looked under {audio_dir})"
                 )
 
             meta: dict[str, Any] = {
@@ -182,8 +179,7 @@ def _load_from_metadata(
 
     if not items:
         raise RuntimeError(
-            f"Deepfake-Eval-2024: no labelled audio loaded from {meta_path} "
-            f"(split={split!r})."
+            f"Deepfake-Eval-2024: no labelled audio loaded from {meta_path} (split={split!r})."
         )
     if missing and skip_missing:
         from detectzoo.utils.logger import get_logger

detectzoo/datasets/audio/for_dataset.py

@@ -278,15 +278,13 @@ def _load_preprocessed(
         empty = [role for role, n in c.items() if n == 0]
         if empty:
             avail = {
-                sp: cs for sp, cs in per_split_counts.items()
-                if all(v > 0 for v in cs.values())
+                sp: cs for sp, cs in per_split_counts.items() if all(v > 0 for v in cs.values())
             }
             hint = (
                 f" Try `split={next(iter(avail))!r}` instead — that split "
                 f"has both classes ({avail[next(iter(avail))]})."
                 if avail
-                else " No other split has both classes either; the local "
-                "extraction is incomplete."
+                else " No other split has both classes either; the local extraction is incomplete."
             )
             raise RuntimeError(
                 f"FoR ({variant_key}, split={split!r}): the {empty!r} class "

detectzoo/datasets/audio/in_the_wild.py

@@ -140,7 +140,9 @@ def _load_from_metadata_csv(root: Path, meta_path: Path) -> List[DatasetItem]:
             raise ValueError(f"In-The-Wild: empty metadata file {meta_path}")
         field_map = {f.strip().lower(): f for f in reader.fieldnames}
         file_col = field_map.get("file") or field_map.get("filename") or field_map.get("path")
-        label_col = field_map.get("label") or field_map.get("class") or field_map.get("ground_truth")
+        label_col = (
+            field_map.get("label") or field_map.get("class") or field_map.get("ground_truth")
+        )
         if not file_col or not label_col:
             raise ValueError(
                 f"In-The-Wild: {meta_path} must contain file and label columns; "
@@ -183,9 +185,7 @@ def _load_from_metadata_csv(root: Path, meta_path: Path) -> List[DatasetItem]:
 def _load_from_class_dirs(root: Path) -> List[DatasetItem]:
     pairs = _class_dirs(root)
     if not pairs:
-        raise FileNotFoundError(
-            f"In-The-Wild: no real/fake class folders under {root}"
-        )
+        raise FileNotFoundError(f"In-The-Wild: no real/fake class folders under {root}")
     items: List[DatasetItem] = []
     for dir_path, label, role in pairs:
         meta = {"modality": "audio", "class": role}

detectzoo/datasets/base.py

@@ -60,9 +60,7 @@ def load(self) -> List[DatasetItem]:
         return self._items
 
     @staticmethod
-    def _balance_and_truncate(
-        items: List[DatasetItem], max_samples: int
-    ) -> List[DatasetItem]:
+    def _balance_and_truncate(items: List[DatasetItem], max_samples: int) -> List[DatasetItem]:
         """Pick ``max_samples`` items balanced across labels 0 and 1.
 
         Takes ``max_samples // 2`` from each class. If one class is short,

detectzoo/datasets/image/__init__.py

@@ -1,12 +1,12 @@
 """Image-modality datasets for AI-generated image detection."""
 
 from detectzoo.datasets.image.aigcdetect import AIGCDetectDataset
+from detectzoo.datasets.image.chameleon import ChameleonDataset
 from detectzoo.datasets.image.cnn_detection import CNNDetectionDataset
 from detectzoo.datasets.image.drct2m import DRCT2MDataset
+from detectzoo.datasets.image.genimage import GenImageDataset
 from detectzoo.datasets.image.self_synthesis import SelfSynthesisDataset
 from detectzoo.datasets.image.univfd import UnivFDDataset
-from detectzoo.datasets.image.genimage import GenImageDataset
-from detectzoo.datasets.image.chameleon import ChameleonDataset
 
 __all__ = [
     "AIGCDetectDataset",

detectzoo/datasets/image/aigcdetect.py

@@ -8,7 +8,8 @@
 Note:
     In the original PatchCraft / AIGCDetectBenchmark setup, the **training split** is based on
     the CNNSpot/CNNDetection training data (i.e., the ForenSynths-style ProGAN-based training
-    set), while AIGCDetect is primarily used as a large, unified test benchmark across many generators.
+    set), while AIGCDetect is primarily used as a large unified test benchmark across
+    many generators.
 
 GitHub: https://github.com/Ekko-zn/AIGCDetectBenchmark
 ModelScope: ``aemilia/AIGCDetectionBenchmark``
@@ -21,8 +22,8 @@
 from pathlib import Path
 from typing import Any, List, Optional, Sequence, Tuple
 
-from detectzoo.datasets.base import BaseDataset, DatasetItem
 from detectzoo.core.registry import register_dataset
+from detectzoo.datasets.base import BaseDataset, DatasetItem
 
 _MODELSCOPE_AIGCDETECT_DATASET: str = "aemilia/AIGCDetectionBenchmark"
 
@@ -67,11 +68,7 @@ def _partition_layout_ok(parent: Path, folder_name: str) -> bool:
         return False
     try:
         for sub in base.iterdir():
-            if (
-                sub.is_dir()
-                and (sub / "0_real").is_dir()
-                and (sub / "1_fake").is_dir()
-            ):
+            if sub.is_dir() and (sub / "0_real").is_dir() and (sub / "1_fake").is_dir():
                 return True
     except OSError:
         return False
@@ -198,9 +195,7 @@ def ensure_aigcdetect_downloaded(
     if found is not None:
         return found
 
-    raise RuntimeError(
-        "Could not locate AIGCDetectBenchmark after ModelScope download."
-    )
+    raise RuntimeError("Could not locate AIGCDetectBenchmark after ModelScope download.")
 
 
 def resolve_aigcdetect_partition(partition: str) -> Tuple[str, str]:
@@ -219,7 +214,7 @@ class AIGCDetectDataset(BaseDataset):
     Parameters
     ----------
     root : str or Path, optional
-        Directory intended to contain partition folders, or a parent to search. 
+        Directory intended to contain partition folders, or a parent to search.
         When omitted, the default cache directory ``.detectzoo_data/aigcdetect/`` is used.
     partitions : sequence of str, optional
         Partition(s) to load. Each entry may be either a **column** name
@@ -248,7 +243,6 @@ def __init__(
         self._resolved_root: Optional[Path] = None
 
     def _data_root(self) -> Path:
-        from detectzoo.datasets._download import get_cache_dir
 
         if self._resolved_root is not None:
             return self._resolved_root
@@ -290,6 +284,12 @@ def _load_all(self) -> List[DatasetItem]:
                 ):
                     for path in sorted(d.rglob("*")):
                         if path.is_file() and path.suffix.lower() in _IMAGE_EXTS:
-                            items.append(DatasetItem(data=str(path), label=label, metadata={**base_meta, "source": source}))
+                            items.append(
+                                DatasetItem(
+                                    data=str(path),
+                                    label=label,
+                                    metadata={**base_meta, "source": source},
+                                )
+                            )
 
         return items