深皿新モデル（回転させたデータで学習）

app.py

@@ -23,9 +23,10 @@ else:
 
 if __name__ == "__main__":
     interface = Interface(
-        fn=lambda modelName, image: infer(
+        fn=lambda modelName, image, threshold: infer(
             input=np.array(PILImage.open(image)),
             suffix=".png",
+            threshold=float(threshold),
             configPath=Path(CONFIGS_PATH + modelName + ".yaml"),
             weightsPath=Path(MODELS_PATH + modelName + ".ckpt")
         ),
@@ -37,6 +38,7 @@ if __name__ == "__main__":
                     "fukazara_scoop_black",
                     "fukazara_scoop_black_2",
                     "fukazara_scoop_black_3",
+                    "fukazara_scoop_black_4",
                 ],
                 label="Model",
             ),
@@ -45,16 +47,33 @@ if __name__ == "__main__":
                 label="Image",
                 file_count="single",
             ),
+            Textbox(
+                type="text",
+                label="Threshold",
+                value="0.55",
+            ),
         ],
         outputs=[
             Textbox(
                 type="text",
                 label="isAnomaly",
             ),
+            Textbox(
+                type="text",
+                label="Anomaly Map Max",
+            ),
+            Textbox(
+                type="text",
+                label="Anomaly Map",
+            ),
             Image(
                 type="numpy",
                 label="Result",
             ),
+            Image(
+                type="numpy",
+                label="Anomaly Map (with custom threshold)",
+            ),
             Image(
                 type="numpy",
                 label="Input",

configs/fukazara_scoop_black_4.yaml

@@ -0,0 +1,112 @@
+# https://github.com/openvinotoolkit/anomalib/blob/main/src/anomalib/models/patchcore/config.yaml
+
+dataset:
+  name: original
+  format: mvtec
+  path: /workspaces/anomaly-detection/datasets/original
+  task: segmentation
+  category: fukazara_scoop_black_4
+  train_batch_size: 32
+  eval_batch_size: 32
+  num_workers: 8
+  image_size: 256 # dimensions to which images are resized (mandatory)
+  # center_crop: 224 # dimensions to which images are center-cropped after resizing (optional)
+  normalization: imagenet # data distribution to which the images will be normalized: [none, imagenet]
+  transform_config:
+    train: null
+    eval: null
+  test_split_mode: synthetic # 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: synthetic # 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)
+  tiling:
+    apply: false
+    tile_size: null
+    stride: null
+    remove_border_count: 0
+    use_random_tiling: False
+    random_tile_count: 16
+
+model:
+  name: patchcore
+  backbone: wide_resnet50_2
+  pre_trained: true
+  layers:
+    - layer2
+    - layer3
+  coreset_sampling_ratio: 0.001
+  num_neighbors: 9
+  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: True # 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: 0
+  path: /workspaces/anomaly-detection/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: 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: 4
+  max_epochs: 1
+  min_epochs: null
+  max_steps: -1
+  min_steps: null
+  max_time: null
+  limit_train_batches: 0.5
+  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

lightning_inference.py

@@ -12,8 +12,9 @@ from anomalib.data.utils import InputNormalizationMethod, get_transforms
 from anomalib.models import get_model
 from anomalib.utils.callbacks import get_callbacks
 
-from utils.preprocess import fill_background_and_crop
-
+from utils.preprocess import fill_background_and_crop, convert_to_png
+from utils.save_anomaly_map_as_png import save_anomaly_map_as_png
+from utils.tensor_to_list import tensor_to_list
 
 def has_blue_pixels(image_path):
     # 画像を読み込む
@@ -29,6 +30,7 @@ def has_blue_pixels(image_path):
 def infer(
         input: np.ndarray,
         suffix: str,
+        threshold,
         configPath: Path,
         weightsPath: Path):
     """Run inference."""
@@ -69,27 +71,34 @@ def infer(
     input_path = Path(tempfile.mktemp(suffix=suffix))
     cv2.imwrite(str(input_path), input)
 
-    fill_background_and_crop(input_path)
+    png_path = Path(convert_to_png(input_path))
+    fill_background_and_crop(png_path)
 
     # create the dataset
     dataset = InferenceDataset(
-        input_path,
+        png_path,
         image_size=tuple(config.dataset.image_size),
         transform=transform
     )
     dataloader = DataLoader(dataset)
 
     # generate predictions
-    predict = trainer.predict(
+    result = trainer.predict(
         model=model,
         dataloaders=[dataloader],
         return_predictions=True,
     )
 
+    anomaly_map = result[0]["anomaly_maps"].squeeze()
+    anomaly_map_path = "/tmp/" + sessionId + "/tmp/anomaly_map.png"
+    save_anomaly_map_as_png(anomaly_map, anomaly_map_path, threshold)
+    anomaly_map_list = tensor_to_list(anomaly_map)
+
     # read the output image
-    output_path = "/tmp/" + sessionId + "/tmp/" + input_path.name
+    output_path = "/tmp/" + sessionId + "/tmp/" + png_path.name
 
     # Check if the output image has blue pixels
-    has_blue = has_blue_pixels(output_path)
+    anomaly_map_max = anomaly_map.max().item()
+    is_anomaly = anomaly_map.max().item() > threshold
 
-    return has_blue, cv2.imread(str(output_path)), cv2.imread(str(input_path))
+    return is_anomaly, anomaly_map_max, anomaly_map_list, cv2.imread(str(output_path)), cv2.imread(str(anomaly_map_path)), cv2.imread(str(png_path))

models/fukazara_scoop_black_4.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9fe2f79b20d6cd9d039740bc866ccf4735c8f9ea9af27a4cc5c0349d44e516de
+size 140731294

requirements.txt

@@ -1,3 +1,3 @@
 openvino==2023.2.0
-anomalib
+anomalib==0.7.0
 wandb

utils/preprocess.py

@@ -7,6 +7,14 @@ import argparse
 from pathlib import Path
 from concurrent.futures import ThreadPoolExecutor, as_completed
 
+
+def convert_to_png(image_path):
+    image = cv2.imread(str(image_path))
+    png_path = str(image_path).rsplit('.', 1)[0] + '.png'
+    cv2.imwrite(png_path, image)
+    return png_path
+
+
 # Function to fill background and crop the image to make the object fill the frame
 def fill_background_and_crop(image_path):
     # 背景色の設定（BGR）