EfficientAD

src/anomalib/data/utils/download.py

@@ -218,6 +218,28 @@ def hash_check(file_path: Path, expected_hash: str) -> None:
         ), f"Downloaded file {file_path} does not match the required hash."
 
 
+def extract(file_name: Path, root: Path) -> None:
+    """Extract a dataset
+
+    Args:
+        file_name (Path): Path of the file to be extracted.
+        root (Path): Root directory where the dataset will be stored.
+
+    """
+    logger.info("Extracting dataset into root folder.")
+    if file_name.suffix == ".zip":
+        with ZipFile(file_name, "r") as zip_file:
+            zip_file.extractall(root)
+    elif file_name.suffix in (".tar", ".gz", ".xz", ".tgz"):
+        with tarfile.open(file_name) as tar_file:
+            safe_extract(tar_file, root)
+    else:
+        raise ValueError(f"Unrecognized file format: {file_name}")
+
+    logger.info("Cleaning up files.")
+    (file_name).unlink()
+
+
 def download_and_extract(root: Path, info: DownloadInfo) -> None:
     """Download and extract a dataset.
 
@@ -246,18 +268,7 @@ def download_and_extract(root: Path, info: DownloadInfo) -> None:
         logger.info("Checking the hash of the downloaded file.")
         hash_check(downloaded_file_path, info.hash)
 
-    logger.info("Extracting dataset into root folder.")
-    if downloaded_file_path.suffix == ".zip":
-        with ZipFile(downloaded_file_path, "r") as zip_file:
-            zip_file.extractall(root)
-    elif downloaded_file_path.suffix in (".tar", ".gz", ".xz"):
-        with tarfile.open(downloaded_file_path) as tar_file:
-            safe_extract(tar_file, root)
-    else:
-        raise ValueError(f"Unrecognized file format: {downloaded_file_path}")
-
-    logger.info("Cleaning up files.")
-    (downloaded_file_path).unlink()
+    extract(downloaded_file_path, root)
 
 
 def is_within_directory(directory: Path, target: Path):

src/anomalib/models/__init__.py

@@ -20,6 +20,7 @@
 from anomalib.models.dfkde import Dfkde
 from anomalib.models.dfm import Dfm
 from anomalib.models.draem import Draem
+from anomalib.models.efficientad import EfficientAD
 from anomalib.models.fastflow import Fastflow
 from anomalib.models.ganomaly import Ganomaly
 from anomalib.models.padim import Padim
@@ -43,6 +44,7 @@
     "Rkde",
     "Stfpm",
     "AiVad",
+    "EfficientAD",
 ]
 
 logger = logging.getLogger(__name__)
@@ -95,6 +97,7 @@ def get_model(config: DictConfig | ListConfig) -> AnomalyModule:
         "rkde",
         "stfpm",
         "ai_vad",
+        "efficientad",
     ]
     model: AnomalyModule
 

src/anomalib/models/efficientad/README.md

@@ -0,0 +1,37 @@
+# EfficientAD
+This is the implementation of the [EfficientAD](https://arxiv.org/pdf/2303.14535.pdf) paper. It is based on https://github.com/rximg/EfficientAD and https://github.com/nelson1425/EfficientAD/
+
+Model Type: Segmentation
+
+## Description
+
+Fast anomaly segmentation algorithm that consists of a distilled pre-trained teacher model, a student model and an autoencoder. It detects local anomalies via the teacher-student discrepany and global anomalies via the student-autoencoder discrepancy.
+
+### Feature Extraction
+Features are extracted from a pre-trained teacher model and used to train a student model and an autoencoder model. To hinder the student from imitating the teacher on anomalies, Imagenet images are used in the loss function.
+
+
+### Anomaly Detection
+Anomalies are detected as the difference in output feature maps between the student model and the autoencoder model.
+
+## Usage
+
+`python tools/train.py --model efficientad`
+
+## Benchmark
+
+All results gathered with seed `42`.
+
+## [MVTec AD Dataset](https://www.mvtec.com/company/research/datasets/mvtec-ad)
+
+### Image-Level AUC
+
+|                          |  Avg  | Carpet | Grid  | Leather | Tile  | Wood  | Bottle | Cable | Capsule | Hazelnut | Metal Nut | Pill  | Screw | Toothbrush | Transistor | Zipper |
+| ------------------------ | :---: | :----: | :---: | :-----: | :---: | :---: | :----: | :---: | :-----: | :------: | :-------: | :---: | :---: | :--------: | :--------: | :----: |
+| Distilled Teacher Medium | 0.000 | 0.000  | 0.000 |  0.000  | 0.000 | 0.000 | 0.000  | 0.000 |  0.000  |  0.000   |   0.000   | 0.000 | 0.000 |   0.000    |   0.000    | 0.000  |
+
+### Image F1 Score
+
+|                          |  Avg  | Carpet | Grid  | Leather | Tile  | Wood  | Bottle | Cable | Capsule | Hazelnut | Metal Nut | Pill  | Screw | Toothbrush | Transistor | Zipper |
+| ------------------------ | :---: | :----: | :---: | :-----: | :---: | :---: | :----: | :---: | :-----: | :------: | :-------: | :---: | :---: | :--------: | :--------: | :----: |
+| Distilled Teacher Medium | 0.000 | 0.000  | 0.000 |  0.000  | 0.000 | 0.000 | 0.000  | 0.000 |  0.000  |  0.000   |   0.000   | 0.000 | 0.000 |   0.000    |   0.000    | 0.000  |

src/anomalib/models/efficientad/__init__.py

@@ -0,0 +1,10 @@
+"""EfficientAD: Accurate Visual Anomaly Detection at Millisecond-Level Latencies.
+https://arxiv.org/pdf/2303.14535.pdf
+"""
+
+# Copyright (C) 2023 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+
+from .lightning_model import EfficientAD, EfficientadLightning
+
+__all__ = ["EfficientAD", "EfficientadLightning"]

src/anomalib/models/efficientad/config.yaml

@@ -0,0 +1,101 @@
+dataset:
+  name: mvtec
+  format: mvtec
+  path: ./datasets/MVTec
+  category: bottle
+  task: segmentation
+  train_batch_size: 1
+  eval_batch_size: 16
+  num_workers: 8
+  image_size: 256 # dimensions to which images are resized (mandatory)
+  center_crop: null # dimensions to which images are center-cropped after resizing (optional)
+  normalization: none # data distribution to which the images will be normalized: [none, imagenet]
+  transform_config:
+    train: null
+    eval: null
+  test_split_mode: from_dir # options: [from_dir, synthetic]
+  test_split_ratio: 0.2 # fraction of train images held out testing (usage depends on test_split_mode)
+  val_split_mode: same_as_test # options: [same_as_test, from_test, synthetic]
+  val_split_ratio: 0.5 # fraction of train/test images held out for validation (usage depends on val_split_mode)
+
+model:
+  name: efficientad
+  teacher_out_channels: 384
+  model_size: medium # options: [small, medium]
+  lr: 0.0001
+  weight_decay: 0.00001
+  padding: true
+  # generic params
+  normalization_method: min_max # options: [null, min_max, cdf]
+
+metrics:
+  image:
+    - F1Score
+    - AUROC
+  pixel:
+    - F1Score
+    - AUROC
+  threshold:
+    method: adaptive #options: [adaptive, manual]
+    manual_image: null
+    manual_pixel: null
+
+visualization:
+  show_images: False # show images on the screen
+  save_images: True # save images to the file system
+  log_images: False # log images to the available loggers (if any)
+  image_save_path: null # path to which images will be saved
+  mode: full # options: ["full", "simple"]
+
+project:
+  seed: 42
+  path: ./results
+
+logging:
+  logger: [] # options: [comet, tensorboard, wandb, csv] or combinations.
+  log_graph: false # Logs the model graph to respective logger.
+
+optimization:
+  export_mode: null # options: torch, onnx, openvino
+# PL Trainer Args. Don't add extra parameter here.
+trainer:
+  enable_checkpointing: true
+  default_root_dir: null
+  gradient_clip_val: 0
+  gradient_clip_algorithm: norm
+  num_nodes: 1
+  devices: 1
+  enable_progress_bar: true
+  overfit_batches: 0.0
+  track_grad_norm: -1
+  check_val_every_n_epoch: 1 # Don't validate before extracting features.
+  fast_dev_run: false
+  accumulate_grad_batches: 1
+  max_epochs: 200
+  min_epochs: null
+  max_steps: -1
+  min_steps: null
+  max_time: null
+  limit_train_batches: 1.0
+  limit_val_batches: 1.0
+  limit_test_batches: 1.0
+  limit_predict_batches: 1.0
+  val_check_interval: 1.0 # Don't validate before extracting features.
+  log_every_n_steps: 50
+  accelerator: auto # <"cpu", "gpu", "tpu", "ipu", "hpu", "auto">
+  strategy: null
+  sync_batchnorm: false
+  precision: 32
+  enable_model_summary: true
+  num_sanity_val_steps: 0
+  profiler: null
+  benchmark: false
+  deterministic: false
+  reload_dataloaders_every_n_epochs: 0
+  auto_lr_find: false
+  replace_sampler_ddp: true
+  detect_anomaly: false
+  auto_scale_batch_size: false
+  plugins: null
+  move_metrics_to_cpu: false
+  multiple_trainloader_mode: max_size_cycle