Refactor dataset_preparation directory using git mv to preserve file …

.gitignore

@@ -7,10 +7,8 @@ __pycache__/
 
 # Outputs
 output/
-outputs/
-dataset/
-assets/classification_dataset
-assets/output_images/
+data/classification_dataset
+data/output_images/
 database/data/
 
 # Virtual environments
@@ -29,4 +27,8 @@ deepstream_engines/
 .ssh/
 .vscode-server/
 .bash_history
-.gitconfig
+.gitconfig
+
+# Dataset - exclude data files
+dataset/data/
+dataset/archives/*.zip

src/optimal_class_mapping.py → dataset/optimal_class_mapping.py

@@ -170,9 +170,7 @@
 }
 
 # Create reverse lookup for dataset names
-_DATASET_NAME_TO_ID: Dict[str, int] = {
-    name: class_id for class_id, name in DATASET_NAMES.items()
-}
+_DATASET_NAME_TO_ID: Dict[str, int] = {name: class_id for class_id, name in DATASET_NAMES.items()}
 
 
 def map_prediction(model_class_id: int) -> int:

dataset_preparation/compress_dataset.py → dataset/preparation/compress_dataset.py

@@ -3,7 +3,9 @@
 from pathlib import Path
 
 
-def compress_dataset(dataset_path="assets/classification_dataset", output_path="classification_dataset.zip"):
+def compress_dataset(
+    dataset_path="data/classification_dataset", output_path="dataset/archives/classification_dataset.zip"
+):
     """
     Compress the classification dataset into a ZIP archive.
 

dataset_preparation/decompress_dataset.py → dataset/preparation/decompress_dataset.py

@@ -2,7 +2,7 @@
 from pathlib import Path
 
 
-def decompress_dataset(archive_path="classification_dataset.zip", output_dir="assets"):
+def decompress_dataset(archive_path="dataset/archives/classification_dataset.zip", output_dir="data"):
     """
     Decompress the classification dataset from a ZIP archive.
 

dataset_preparation/prepare_crops_dataset.py → dataset/preparation/prepare_crops_dataset.py

@@ -170,7 +170,7 @@ def create_largest_k_classification_dataset(
     """
     images_dir = Path(data_root) / "images"
     bboxes_dir = Path(data_root) / "bboxes" / "CropAndWeed"
-    output_dir = Path("assets") / "classification_dataset" / "images"
+    output_dir = Path("data") / "classification_dataset" / "images"
     output_dir.mkdir(parents=True, exist_ok=True)
 
     image_files = {img_file.stem: img_file for img_file in images_dir.glob("*.jpg")}
@@ -188,7 +188,7 @@ def create_largest_k_classification_dataset(
     print("\n✂️ Creating dataset with LARGEST objects (Largest-K strategy)...")
     stats, total_objects = save_largest_k_crops(eligible_classes, output_dir, max_objects_per_class)
 
-    stats_file = Path("assets") / "classification_dataset" / "dataset_stats_largest_k.json"
+    stats_file = Path("data") / "classification_dataset" / "dataset_stats_largest_k.json"
     with open(stats_file, "w") as f:
         json.dump(stats, f, indent=2)
 

docs/dataset_preparation.md

@@ -16,29 +16,29 @@ This project uses a **two-stage dataset preparation process**:
 
 Download the original CropAndWeed dataset (5 tar archives, ~10,000 images) from the official source:
 ```bash
-python dataset_preparation/dataset_prep.py --data_root ./dataset
+python dataset/preparation/dataset_prep.py --data_root ./dataset/data
 ```
 - **What gets downloaded:** Five large tar files containing images and annotations.
 - **Expected download time:** Each archive is several GB; total time depends on your internet speed (typically 10–30 minutes).
 - **Resulting structure:**  
   ```
-  dataset/
-  ├── images/
-  ├── bboxes/
-  ├── labelIds/
-  └── params/
+dataset/data/
+├── images/
+├── bboxes/
+├── labelIds/
+└── params/
   ```
 
 ### Stage 2: Create Balanced Classification Dataset
 
 Generate a balanced classification dataset (76 classes, 12 objects per class, 912 total) from the original data:
 ```bash
-python dataset_preparation/prepare_crops_dataset.py --data_root ./dataset
+python dataset/preparation/prepare_crops_dataset.py --data_root ./dataset/data
 ```
 - **What happens:** The script filters and crops the original images using **PIL** to create a balanced classification dataset.
 - **Output:**  
   ```
-  classification_dataset/
+  data/classification_dataset/
   ├── images/
   │   ├── 1/
   │   ├── 2/
@@ -50,7 +50,7 @@ python dataset_preparation/prepare_crops_dataset.py --data_root ./dataset
 
 If you only need the ready-to-use classification dataset, extract it directly:
 ```bash
-python dataset_preparation/decompress_dataset.py
+python dataset/preparation/decompress_dataset.py
 ```
 - **What happens:** Extracts a pre-built balanced dataset (912 images, 76 classes) for immediate use.
 
@@ -75,7 +75,7 @@ The **CropAndWeed** dataset contains agricultural images with multi-level annota
 
 ### Native Format (Original)
 ```
-dataset/
+dataset/data/
 ├── images/                # RGB images (JPG format)
 │   ├── ave-0000-0001.jpg
 │   └── ...
@@ -163,10 +163,10 @@ To access the ready-to-use classification dataset:
 
 ```bash
 # Extract classification dataset archive (updated with Largest-K)
-python dataset_preparation/decompress_dataset.py
+python dataset/preparation/decompress_dataset.py
 
 # Or extract from custom location
-python dataset_preparation/decompress_dataset.py classification_dataset.zip extracted_data/
+python dataset/preparation/decompress_dataset.py dataset/archives/classification_dataset.zip extracted_data/
 ```
 
 **Verification:** Script automatically verifies 912 images and 2 JSON files.
@@ -185,12 +185,12 @@ The following issues are known and documented as part of the Largest-K approach:
 
 ```bash
 # Generate dataset with Largest-K strategy
-python dataset_preparation/prepare_crops_dataset.py --data_root ./dataset
+python dataset/preparation/prepare_crops_dataset.py --data_root ./dataset/data
 ```
 
 ### Output files:
-- `assets/classification_dataset/images/` - Cropped objects organized by class
-- `assets/classification_dataset/dataset_stats_largest_k.json` - Size distribution statistics
+- `data/classification_dataset/images/` - Cropped objects organized by class
+- `data/classification_dataset/dataset_stats_largest_k.json` - Size distribution statistics
 
 ---
 
@@ -205,4 +205,4 @@ python dataset_preparation/prepare_crops_dataset.py --data_root ./dataset
 - **Full dataset** (~10k images, 85 classes)
 - **Single-class images** (~1.5k images, ~60 classes)
 - **Single-object images** (~500 images, ~40 classes)
-- **Largest-K dataset** (912 images, 76 classes) - **Current approach**
+- **Largest-K dataset** (912 images, 76 classes) - **Current approach**

src/evaluation/run_evaluation.py

@@ -7,7 +7,7 @@
 from src.evaluation.evaluator import EvaluationConfig, ModelEvaluator
 from src.inference import MobileNetInference, ResNetInference
 from src.onnx_model import OnnxClassifierInferenceBase as OnnxModel
-from src.optimal_class_mapping import map_prediction
+from dataset.optimal_class_mapping import map_prediction
 from src.path_utils import ensure_clean_directory
 
 MODELS_DIR_PATH = Path("models")

src/evaluation/run_hierarchical_evaluation.py

@@ -4,11 +4,11 @@
 
 from PIL import Image
 
-from dataset_preparation.utilities.datasets import DATASETS
+from dataset.utilities.datasets import DATASETS
 from metrics.metrics_api import compute_metrics
 from src.evaluation.evaluator import EvaluationConfig, ModelEvaluator
 from src.inference import MobileNetInference, ResNetInference
-from src.optimal_class_mapping import map_prediction
+from dataset.optimal_class_mapping import map_prediction
 
 
 class MappedModelWrapper: