Feat/hierarchical output paths

README.md

@@ -139,7 +139,7 @@ anomavision detect --config config.yml --img_path ./test_images --thresh 13.0
 anomavision eval --config config.yml --enable_visualization
 
 # Export to ONNX / TorchScript / OpenVINO / all
-anomavision export --model padim_model.pt --format all --precision fp16
+anomavision export --config config.yml --model model.pt --format all --precision fp16
 ```
 
 Every command has full `--help`:
@@ -187,8 +187,8 @@ model = anomavision.Padim(
 model.fit(loader)
 
 # --- 3. Save ---
-torch.save(model, "padim_model.pt")                  # full model (for export or further use)
-model.save_statistics("padim_model.pth", half=True)  # stats-only (smaller, faster to load)
+torch.save(model, "model.pt")                  # full model (for export or further use)
+model.save_statistics("model.pth", half=True)  # stats-only (smaller, faster to load)
 
 # --- 4. Infer ---
 # scores: (batch_size,)      — scalar anomaly score per image. Higher = more anomalous.
@@ -302,7 +302,7 @@ Full docs at **http://localhost:8000/docs** once the server is running.
 ```bash
 anomavision export \
   --model_data_path ./distributions/anomav_exp \
-  --model padim_model.pt \
+  --model model.pt \
   --format onnx \
   --precision fp16 \
   --quantize-dynamic
@@ -329,7 +329,7 @@ stream_mode: true
 stream_source:
   type: webcam
   camera_id: 0
-model: padim_model.onnx
+model: model.onnx
 thresh: 13.0
 enable_visualization: true
 ```
@@ -361,11 +361,11 @@ backbone:        resnet18
 batch_size:      16
 feat_dim:        100
 layer_indices:   [0, 1, 2]
-output_model:    padim_model.pt
+output_model:    model.pt
 run_name:        exp1
 model_data_path: ./distributions/anomav_exp
 
-model:           padim_model.onnx
+model:           model.onnx
 device:          auto        # auto | cpu | cuda
 thresh:          13.0
 

anomavision/cli.py

@@ -11,9 +11,9 @@
 
 Examples:
     anomavision train --config config.yml
-    anomavision export --model padim_model.pt --format onnx
-    anomavision detect --model padim_model.onnx --img_path ./test_images
-    anomavision eval --model padim_model.pt --class_name bottle
+    anomavision export --config config.yml --model model.pt --format onnx
+    anomavision detect --config config.yml --model model.onnx --img_path ./test_images
+    anomavision eval --config config.yml --model model.pt --class_name bottle
 """
 
 import argparse
@@ -35,9 +35,9 @@ def create_parser() -> argparse.ArgumentParser:
         epilog="""
 Examples:
   %(prog)s train --config config.yml --dataset_path /data --class_name bottle
-  %(prog)s export --model padim_model.pt --format onnx --quantize-dynamic
-  %(prog)s detect --model padim_model.onnx --img_path ./test --enable_visualization
-  %(prog)s eval --model padim_model.pt --class_name bottle --dataset_path /data
+  %(prog)s export --model model.pt --format onnx --quantize-dynamic
+  %(prog)s detect --model model.onnx --img_path ./test --enable_visualization
+  %(prog)s eval --model model.pt --class_name bottle --dataset_path /data
 
 For detailed help on each command:
   %(prog)s train --help

anomavision/detect.py

@@ -1,11 +1,11 @@
 """
 Run Anomaly detection inference on images using various model formats.
 Usage - formats:
-    $ python detect.py --model padim_model.pt                  # PyTorch
-                                   padim_model.torchscript        # TorchScript
-                                   padim_model.onnx               # ONNX Runtime
-                                   padim_model_openvino           # OpenVINO
-                                   padim_model.engine             # TensorRT
+    $ python detect.py --model model.pt                     # PyTorch
+                                   model.torchscript        # TorchScript
+                                   model.onnx               # ONNX Runtime
+                                   model_openvino           # OpenVINO
+                                   model.engine             # TensorRT
 """
 
 import argparse
@@ -61,13 +61,19 @@ def create_parser(add_help: bool = True) -> argparse.ArgumentParser:
     parser.add_argument(
         "--model_data_path",
         type=str,
-        default="./distributions/anomav_exp",
+        default="./distributions",
         help="Directory containing model files.",
     )
+    parser.add_argument(
+        "--algorithm",
+        type=str,
+        default=None,
+        help="Algorithm name (e.g., padim, patchcore).",
+    )
     parser.add_argument(
         "--model",
         type=str,
-        default="padim_model.pt",
+        default=None,
         help="Model file (.pt for PyTorch, .onnx for ONNX, .engine for TensorRT)",
     )
     parser.add_argument(
@@ -121,7 +127,7 @@ def create_parser(add_help: bool = True) -> argparse.ArgumentParser:
     )
     parser.add_argument(
         "--run_name",
-        default="detect_exp",
+        default=None,
         help="experiment name for this inference run",
     )
     parser.add_argument(
@@ -178,7 +184,19 @@ def run_inference(args):
         cfg = load_config(str(args.config))
     else:
         # Fallback to model directory config
-        cfg = load_config(str(Path(args.model_data_path) / "config.yml"))
+        potential_paths = []
+        if args.model_data_path:
+            base_path = Path(args.model_data_path)
+            potential_paths.append(base_path / "config.yml")
+
+        cfg = {}
+        for path in potential_paths:
+            if path.exists():
+                cfg = load_config(str(path))
+                break
+
+        if not cfg:
+            cfg = {}
 
     # Merge config with CLI args
     config = edict(merge_config(args, cfg))
@@ -253,7 +271,7 @@ def run_inference(args):
 
     # --- Model Loading Phase ---
     with profilers["model_loading"]:
-        model_path = os.path.join(MODEL_DATA_PATH, config.model)
+        model_path = os.path.join(MODEL_DATA_PATH, config.algorithm, config.class_name, config.run_name, config.model)
         logger.info(f"Loading model: {model_path}")
 
         if not os.path.exists(model_path):
@@ -273,7 +291,7 @@ def run_inference(args):
         run_name = config.run_name
         viz_output_dir = config.get("viz_output_dir", "./visualizations/")
         RESULTS_PATH = increment_path(
-            Path(viz_output_dir) / model_type.value.upper() / run_name,
+            Path(viz_output_dir) / config.algorithm / config.class_name / model_type.value.upper() / run_name,
             exist_ok=config.get("overwrite", False),
             mkdir=True,
         )

anomavision/eval.py

@@ -21,6 +21,9 @@
     get_logger,
     merge_config,
     setup_logging,
+    find_best_threshold_f1,
+    compute_metrics,
+    find_optimal_threshold
 )
 
 
@@ -40,7 +43,6 @@ def create_parser(add_help: bool = True) -> argparse.ArgumentParser:
         "--dataset_path",
         default=None,
         type=str,
-        required=False,
         help="Path to the dataset folder containing test images.",
     )
     parser.add_argument(
@@ -54,13 +56,19 @@ def create_parser(add_help: bool = True) -> argparse.ArgumentParser:
     parser.add_argument(
         "--model_data_path",
         type=str,
-        default="./distributions/anomav_exp",
+        default=None,
         help="Directory containing AnomaVision model files.",
     )
+    parser.add_argument(
+        "--algorithm",
+        type=str,
+        default=None,
+        help="Algorithm name (e.g., padim, patchcore).",
+    )
     parser.add_argument(
         "--model",
         type=str,
-        default="padim_model.pt",
+        default=None,
         help="Model filename (.pt, .onnx, .engine)",
     )
     parser.add_argument(
@@ -110,7 +118,7 @@ def create_parser(add_help: bool = True) -> argparse.ArgumentParser:
     parser.add_argument(
         "--viz_output_dir",
         type=str,
-        default="./eval_visualizations/",
+        default=None,
         help="Directory to save visualization images.",
     )
 
@@ -131,38 +139,6 @@ def create_parser(add_help: bool = True) -> argparse.ArgumentParser:
     return parser
 
 
-def compute_metrics(labels, scores, thresh=None):
-    """
-    Calculate standard anomaly detection metrics.
-    """
-    metrics = {}
-
-    # AUROC
-    try:
-        metrics['auc_score'] = float(roc_auc_score(labels, scores))
-    except ValueError:
-        metrics['auc_score'] = 0.0
-
-    # PR-AUC
-    try:
-        precision, recall, _ = precision_recall_curve(labels, scores)
-        metrics['pr_auc'] = float(auc(recall, precision))
-    except ValueError:
-        metrics['pr_auc'] = 0.0
-
-    # Statistics
-    metrics['mean_anomaly_score'] = float(np.mean(scores))
-    metrics['std_anomaly_score'] = float(np.std(scores))
-
-    # Accuracy (if thresh provided)
-    if thresh is not None:
-        predictions = (scores > thresh).astype(int)
-        metrics['accuracy'] = float(np.mean(predictions == labels))
-        metrics['thresh'] = thresh
-
-    return metrics
-
-
 def evaluate_model_with_wrapper(
     model_wrapper, test_dataloader, logger, evaluation_profiler, detailed_timing=False
 ):
@@ -263,7 +239,7 @@ def run_evaluation(args):
 
     # Load Model Phase
     with profilers["model_loading"]:
-        model_path = os.path.join(MODEL_DATA_PATH, config.model)
+        model_path = os.path.join(MODEL_DATA_PATH, config.algorithm, config.class_name, config.run_name, config.model)
         logger.info(f"Loading model: {model_path}")
 
         if not os.path.exists(model_path):
@@ -323,7 +299,12 @@ def run_evaluation(args):
         model_wrapper.close()
 
     # Compute Metrics
-    metrics = compute_metrics(labels, scores, thresh=config.thresh)
+    if config.thresh is None:
+        best_thresh, _ = find_optimal_threshold(labels, scores)
+    else:
+        best_thresh = config.thresh
+
+    metrics = compute_metrics(labels, scores, thresh=best_thresh)
 
     # Add timing metrics
     total_images = len(test_dataset)
@@ -368,7 +349,7 @@ def run_evaluation(args):
     logger.info(f"Data loading time:         {profilers['data_loading'].accumulated_time * 1000:.2f} ms")
     logger.info(f"Evaluation time:           {profilers['evaluation'].accumulated_time * 1000:.2f} ms")
     logger.info(f"Visualization time:        {profilers['visualization'].accumulated_time * 1000:.2f} ms")
-    logger.info("=" * 60)
+    # logger.info("=" * 60)
 
     # 2. PERFORMANCE METRICS
     logger.info("=" * 60)
@@ -382,7 +363,7 @@ def run_evaluation(args):
     if len(test_dataloader) > 0:
         images_per_batch = total_images / len(test_dataloader)
         logger.info(f"Evaluation throughput:     {evaluation_fps * images_per_batch:.1f} images/sec (batch size: {batch_size})")
-    logger.info("=" * 60)
+    # logger.info("=" * 60)
 
     # 3. EVALUATION SUMMARY
     logger.info("=" * 60)
@@ -393,6 +374,18 @@ def run_evaluation(args):
     logger.info(f"Model type: {model_type.value.upper() if model_type else 'UNKNOWN'}")
     logger.info(f"Device: {device_str}")
     logger.info(f"Image processing: resize={config.resize}, crop_size={config.crop_size}, normalize={config.normalize}")
+    # logger.info("=" * 60)
+
+    logger.info("=" * 60)
+    logger.info("ANOMAVISION DETECTION METRICS")
+    logger.info("=" * 60)
+
+    for k, v in metrics.items():
+        if isinstance(v, float):
+            logger.info(f"{k.replace('_',' ').title():<28} {v:.6f}")
+        else:
+            logger.info(f"{k.replace('_',' ').title():<28} {v}")
+
     logger.info("=" * 60)
 
     logger.info("AnomaVision anomaly detection model evaluation completed successfully")

anomavision/export.py

@@ -604,10 +604,17 @@ def create_parser(add_help: bool = True) -> argparse.ArgumentParser:
     parser.add_argument(
         "--model_data_path",
         type=str,
-        default="./distributions/anomav_exp",
+        default="./distributions",
         help="Directory containing model and output location",
     )
 
+    parser.add_argument(
+        "--algorithm",
+        type=str,
+        default=None,
+        help="Algorithm name (e.g., padim, patchcore).",
+    )
+
     parser.add_argument(
         "--model",
         type=str,
@@ -692,16 +699,31 @@ def main(args=None):
     if args.config is not None:
         cfg = load_config(str(args.config))
     else:
-        cfg = load_config(str(Path(args.model_data_path) / "config.yml"))
+        # Try to find config in structured path first
+        potential_paths = []
+        if args.model_data_path:
+            # Try to infer from model path if it follows structure
+            base_path = Path(args.model_data_path)
+            potential_paths.append(base_path / "config.yml")
+
+        # Use first existing config
+        cfg = {}
+        for path in potential_paths:
+            if path.exists():
+                cfg = load_config(str(path))
+                break
+
+        if not cfg:
+            cfg = {}
 
     config = edict(merge_config(args, cfg))
 
     # Setup logging & logger
     setup_logging(enabled=True, log_level=config.log_level, log_to_file=True)
     logger = get_logger("anomavision.export")
 
-    model_path = Path(config.model_data_path) / config.model
-    output_dir = Path(config.model_data_path)
+    model_path = Path(config.model_data_path) / config.algorithm / config.class_name / config.run_name / config.model
+    output_dir = Path(config.model_data_path) / config.algorithm / config.class_name / config.run_name
     model_stem = Path(config.model).stem
 
     # Generate output names

anomavision/train.py

@@ -124,6 +124,12 @@ def create_parser(add_help: bool = True) -> argparse.ArgumentParser:
         default=None,
         help="Directory to save model distributions and PT file.",
     )
+    parser.add_argument(
+        "--algorithm",
+        type=str,
+        default=None,
+        help="Algorithm name (e.g., padim, patchcore).",
+    )
     parser.add_argument(
         "--log_level",
         type=str,
@@ -176,7 +182,7 @@ def run_training(args):
 
     # Resolve output run dir once
     run_dir = increment_path(
-        Path(config.model_data_path) / config.run_name, exist_ok=True, mkdir=True
+        Path(config.model_data_path) / config.algorithm / config.class_name / config.run_name, exist_ok=True, mkdir=True
     )
 
     # === Dataset ===