geoai-py 0.26.0__py2.py3-none-any.whl → 0.27.0__py2.py3-none-any.whl

geoai/__init__.py

@@ -2,7 +2,7 @@
 
 __author__ = """Qiusheng Wu"""
 __email__ = "giswqs@gmail.com"
-__version__ = "0.26.0"
+__version__ = "0.27.0"
 
 
 import os
@@ -101,6 +101,12 @@ def set_proj_lib_path(verbose=False):
 
 from .dinov3 import DINOv3GeoProcessor, analyze_image_patches, create_similarity_map
 from .geoai import *
+from .utils import (
+    orthogonalize,
+    regularization,
+    hybrid_regularization,
+    adaptive_regularization,
+)
 
 from .timm_train import (
     get_timm_model,
@@ -122,6 +128,25 @@ from .timm_segment import (
     push_timm_model_to_hub,
 )
 
+from .timm_regress import (
+    PixelRegressionModel,
+    PixelRegressionDataset,
+    create_regression_tiles,
+    train_pixel_regressor,
+    predict_raster,
+    evaluate_regression,
+    plot_regression_comparison,
+    plot_scatter,
+    plot_training_history,
+    visualize_prediction,
+    plot_regression_results,
+    # Backward compatibility aliases
+    TimmRegressor,
+    RegressionDataset,
+    train_timm_regressor,
+    create_regression_patches,
+)
+
 # Import tools subpackage
 from . import tools
 
@@ -167,6 +192,7 @@ except ImportError:
 try:
     from .prithvi import (
         PrithviProcessor,
+        get_available_prithvi_models,
         load_prithvi_model,
         prithvi_inference,
     )

geoai/auto.py

@@ -24,7 +24,6 @@ Example:
 import os
 from typing import Any, Dict, List, Optional, Tuple, Union
 
-import cv2
 import geopandas as gpd
 import numpy as np
 import rasterio
@@ -877,6 +876,8 @@ class AutoGeoModel:
         **kwargs: Any,
     ) -> Dict[str, Any]:
         """Run tiled inference for large images."""
+        import cv2  # Lazy import to avoid QGIS opencv conflicts
+
         if data.ndim == 3:
             _, height, width = data.shape
         else:
@@ -1872,6 +1873,7 @@ def show_segmentation(
         >>> result = geoai.auto.semantic_segmentation("aerial.tif", output_path="seg.tif")
         >>> fig = show_segmentation("aerial.tif", result["mask"])
     """
+    import cv2  # Lazy import to avoid QGIS opencv conflicts
     import matplotlib.pyplot as plt
 
     img, _ = _load_image_for_display(source)
@@ -1941,6 +1943,7 @@ def show_depth(
         >>> result = geoai.auto.depth_estimation("aerial.tif", output_path="depth.tif")
         >>> fig = show_depth("aerial.tif", result["depth"])
     """
+    import cv2  # Lazy import to avoid QGIS opencv conflicts
     import matplotlib.pyplot as plt
 
     img, _ = _load_image_for_display(source)

geoai/change_detection.py

@@ -3,7 +3,6 @@
 import os
 from typing import Any, Dict, List, Optional, Tuple, Union
 
-import cv2
 import matplotlib.pyplot as plt
 import numpy as np
 import rasterio
@@ -736,6 +735,7 @@ class ChangeDetection:
         output_path="split_comparison.png",
     ):
         """Create a split comparison visualization showing before/after with change overlay."""
+        import cv2  # Lazy import to avoid QGIS opencv conflicts
 
         # Load data
         with rasterio.open(image1_path) as src:

geoai/detectron2.py

@@ -6,7 +6,6 @@ import os
 import warnings
 from typing import Dict, List, Optional, Tuple, Union
 
-import cv2
 import numpy as np
 import rasterio
 import torch
@@ -135,6 +134,8 @@ def detectron2_segment(
     Returns:
         Dict containing segmentation results and output file paths
     """
+    import cv2  # Lazy import to avoid QGIS opencv conflicts
+
     check_detectron2()
 
     # Load the model
@@ -315,6 +316,8 @@ def visualize_detectron2_results(
     Returns:
         Visualization image as numpy array
     """
+    import cv2  # Lazy import to avoid QGIS opencv conflicts
+
     check_detectron2()
 
     # Load the image

geoai/extract.py

@@ -6,7 +6,6 @@ import time
 from typing import Any, Dict, Generator, List, Optional, Tuple, Union
 
 # Third-Party Libraries
-import cv2
 import geopandas as gpd
 import matplotlib.pyplot as plt
 import numpy as np
@@ -440,6 +439,7 @@ class ObjectDetector:
         Returns:
             List of polygons as lists of (x, y) coordinates
         """
+        import cv2  # Lazy import to avoid QGIS opencv conflicts
 
         # Get parameters from kwargs or use instance defaults
         simplify_tolerance = kwargs.get("simplify_tolerance", self.simplify_tolerance)
@@ -637,6 +637,8 @@ class ObjectDetector:
         Returns:
             GeoDataFrame with objects
         """
+        import cv2  # Lazy import to avoid QGIS opencv conflicts
+
         # Use class defaults if parameters not provided
         simplify_tolerance = (
             simplify_tolerance
@@ -2165,6 +2167,7 @@ class ObjectDetector:
         Returns:
             GeoDataFrame with car detections and confidence values
         """
+        import cv2  # Lazy import to avoid QGIS opencv conflicts
 
         def _process_single_component(
             component_mask: np.ndarray,

geoai/prithvi.py

@@ -28,6 +28,16 @@ NO_DATA_FLOAT = 0.0001
 OFFSET = 0
 PERCENTILE = 99.9
 
+# Available Prithvi models
+AVAILABLE_MODELS = [
+    "Prithvi-EO-2.0-tiny-TL",  # tiny transfer learning, embed_dim=192, depth=12, with coords
+    "Prithvi-EO-2.0-100M-TL",  # 100M transfer learning, embed_dim=768, depth=12, with coords
+    "Prithvi-EO-2.0-300M",  # 300M base model, embed_dim=1024, depth=24, no coords
+    "Prithvi-EO-2.0-300M-TL",  # 300M transfer learning, embed_dim=768, depth=12, with coords
+    "Prithvi-EO-2.0-600M",  # 600M base model, embed_dim=1280, depth=32, no coords
+    "Prithvi-EO-2.0-600M-TL",  # 600M transfer learning, embed_dim=1280, depth=32, with coords
+]
+
 
 def get_3d_sincos_pos_embed(embed_dim, grid_size, add_cls_token=False):
     """Create 3D sin/cos positional embeddings.
@@ -622,8 +632,22 @@ class PrithviMAE(nn.Module):
 class PrithviProcessor:
     """Prithvi EO 2.0 processor with GeoTIFF input/output support.
 
-    https://huggingface.co/ibm-nasa-geospatial/Prithvi-EO-2.0-300M-TL
-    https://github.com/NASA-IMPACT/Prithvi-EO-2.0
+    Supports multiple model variants:
+    - Prithvi-EO-2.0-tiny-TL (tiny transfer learning)
+    - Prithvi-EO-2.0-100M-TL (100M transfer learning)
+    - Prithvi-EO-2.0-300M (300M base model)
+    - Prithvi-EO-2.0-300M-TL (300M transfer learning)
+    - Prithvi-EO-2.0-600M (600M base model)
+    - Prithvi-EO-2.0-600M-TL (600M transfer learning)
+
+    References:
+        - tiny-TL: https://huggingface.co/ibm-nasa-geospatial/Prithvi-EO-2.0-tiny-TL
+        - 100M-TL: https://huggingface.co/ibm-nasa-geospatial/Prithvi-EO-2.0-100M-TL
+        - 300M: https://huggingface.co/ibm-nasa-geospatial/Prithvi-EO-2.0-300M
+        - 300M-TL: https://huggingface.co/ibm-nasa-geospatial/Prithvi-EO-2.0-300M-TL
+        - 600M: https://huggingface.co/ibm-nasa-geospatial/Prithvi-EO-2.0-600M
+        - 600M-TL: https://huggingface.co/ibm-nasa-geospatial/Prithvi-EO-2.0-600M-TL
+        - GitHub: https://github.com/NASA-IMPACT/Prithvi-EO-2.0
     """
 
     def __init__(
@@ -637,7 +661,14 @@ class PrithviProcessor:
         """Initialize Prithvi processor.
 
         Args:
-            model_name: Name of the Prithvi model to download from HuggingFace Hub
+            model_name: Name of the Prithvi model to download from HuggingFace Hub.
+                Options:
+                - "Prithvi-EO-2.0-tiny-TL" (tiny, 192 dim, 12 layers)
+                - "Prithvi-EO-2.0-100M-TL" (100M, 768 dim, 12 layers)
+                - "Prithvi-EO-2.0-300M" (base, 1024 dim, 24 layers)
+                - "Prithvi-EO-2.0-300M-TL" (default, 768 dim, 12 layers)
+                - "Prithvi-EO-2.0-600M" (base, 1280 dim, 32 layers)
+                - "Prithvi-EO-2.0-600M-TL" (1280 dim, 32 layers)
             config_path: Path to config file (optional, downloads if not provided)
             checkpoint_path: Path to checkpoint file (optional, downloads if not provided)
             device: Torch device to use
@@ -679,7 +710,13 @@ class PrithviProcessor:
         """Download Prithvi model from HuggingFace Hub.
 
         Args:
-            model_name: Name of the model
+            model_name: Name of the model. Options:
+                - "Prithvi-EO-2.0-tiny-TL"
+                - "Prithvi-EO-2.0-100M-TL"
+                - "Prithvi-EO-2.0-300M" (base model)
+                - "Prithvi-EO-2.0-300M-TL" (default)
+                - "Prithvi-EO-2.0-600M" (base model)
+                - "Prithvi-EO-2.0-600M-TL"
             cache_dir: Directory to cache files
 
         Returns:
@@ -1208,6 +1245,20 @@ class PrithviProcessor:
                 dest.write(image[i], i + 1)
 
 
+def get_available_prithvi_models() -> List[str]:
+    """Get list of available Prithvi model names.
+
+    Returns:
+        List of available model names
+
+    Example:
+        >>> models = get_available_prithvi_models()
+        >>> print(models)
+        ['Prithvi-EO-2.0-300M-TL', 'Prithvi-EO-2.0-600M-TL']
+    """
+    return AVAILABLE_MODELS.copy()
+
+
 def load_prithvi_model(
     model_name: str = "Prithvi-EO-2.0-300M-TL",
     device: Optional[str] = None,
@@ -1216,12 +1267,32 @@ def load_prithvi_model(
     """Load Prithvi model (convenience function).
 
     Args:
-        model_name: Name of the model
+        model_name: Name of the model. Options:
+            - "Prithvi-EO-2.0-tiny-TL"
+            - "Prithvi-EO-2.0-100M-TL"
+            - "Prithvi-EO-2.0-300M" (base)
+            - "Prithvi-EO-2.0-300M-TL" (default)
+            - "Prithvi-EO-2.0-600M" (base)
+            - "Prithvi-EO-2.0-600M-TL"
         device: Device to use ('cuda' or 'cpu')
         cache_dir: Cache directory
 
     Returns:
         PrithviProcessor instance
+
+    Example:
+        >>> # Load tiny-TL model
+        >>> processor = load_prithvi_model("Prithvi-EO-2.0-tiny-TL")
+        >>> # Load 100M-TL model
+        >>> processor = load_prithvi_model("Prithvi-EO-2.0-100M-TL")
+        >>> # Load 300M base model
+        >>> processor = load_prithvi_model("Prithvi-EO-2.0-300M")
+        >>> # Load 300M-TL model
+        >>> processor = load_prithvi_model("Prithvi-EO-2.0-300M-TL")
+        >>> # Load 600M base model
+        >>> processor = load_prithvi_model("Prithvi-EO-2.0-600M")
+        >>> # Load 600M-TL model
+        >>> processor = load_prithvi_model("Prithvi-EO-2.0-600M-TL")
     """
     if device is not None:
         device = torch.device(device)
@@ -1245,9 +1316,23 @@ def prithvi_inference(
     Args:
         file_paths: List of input GeoTIFF files
         output_dir: Output directory
-        model_name: Name of the model
+        model_name: Name of the model. Options:
+            - "Prithvi-EO-2.0-tiny-TL"
+            - "Prithvi-EO-2.0-100M-TL"
+            - "Prithvi-EO-2.0-300M" (base)
+            - "Prithvi-EO-2.0-300M-TL" (default)
+            - "Prithvi-EO-2.0-600M" (base)
+            - "Prithvi-EO-2.0-600M-TL"
         mask_ratio: Optional mask ratio
         device: Device to use
+
+    Example:
+        >>> # Use tiny-TL model
+        >>> prithvi_inference(
+        ...     file_paths=["img1.tif", "img2.tif", "img3.tif", "img4.tif"],
+        ...     model_name="Prithvi-EO-2.0-tiny-TL",
+        ...     output_dir="output_tiny"
+        ... )
     """
     processor = load_prithvi_model(model_name, device)
     processor.process_files(file_paths, output_dir, mask_ratio)

geoai/sam.py

@@ -5,7 +5,6 @@ The SamGeo class provides an interface for segmenting geospatial data using the
 import os
 from typing import Any, Dict, List, Optional, Tuple, Union
 
-import cv2
 import numpy as np
 import torch
 from leafmap import array_to_image, blend_images
@@ -125,6 +124,7 @@ class SamGeo:
         Raises:
             ValueError: If the input source is not a valid path or numpy array.
         """
+        import cv2  # Lazy import to avoid QGIS opencv conflicts
 
         if isinstance(source, str):
             if source.startswith("http"):
@@ -399,6 +399,7 @@ class SamGeo:
         Raises:
             ValueError: If no masks are available and `save_masks()` cannot generate them.
         """
+        import cv2  # Lazy import to avoid QGIS opencv conflicts
         import matplotlib.pyplot as plt
 
         if self.batch:

geoai/segment.py

@@ -4,7 +4,6 @@ import os
 from dataclasses import dataclass
 from typing import Any, Dict, List, Optional, Tuple, Union
 
-import cv2
 import geopandas as gpd
 import numpy as np
 import rasterio
@@ -174,6 +173,8 @@ class GroundedSAM:
         Returns:
             List[DetectionResult]: Filtered detection results.
         """
+        import cv2  # Lazy import to avoid QGIS opencv conflicts
+
         if not detections:
             return detections
 
@@ -235,6 +236,8 @@ class GroundedSAM:
 
     def _mask_to_polygon(self, mask: np.ndarray) -> List[List[int]]:
         """Convert mask to polygon coordinates."""
+        import cv2  # Lazy import to avoid QGIS opencv conflicts
+
         # Find contours in the binary mask
         contours, _ = cv2.findContours(
             mask.astype(np.uint8), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE
@@ -255,6 +258,8 @@ class GroundedSAM:
         self, polygon: List[Tuple[int, int]], image_shape: Tuple[int, int]
     ) -> np.ndarray:
         """Convert polygon to mask."""
+        import cv2  # Lazy import to avoid QGIS opencv conflicts
+
         # Create an empty mask
         mask = np.zeros(image_shape, dtype=np.uint8)
 
@@ -279,6 +284,8 @@ class GroundedSAM:
         Returns:
             List[np.ndarray]: List of individual instance masks.
         """
+        import cv2  # Lazy import to avoid QGIS opencv conflicts
+
         # Find connected components
         num_labels, labels, stats, centroids = cv2.connectedComponentsWithStats(
             mask.astype(np.uint8), connectivity=8
@@ -320,6 +327,8 @@ class GroundedSAM:
         Returns:
             List[Dict]: List of polygon dictionaries with geometry and properties.
         """
+        import cv2  # Lazy import to avoid QGIS opencv conflicts
+
         polygons = []
 
         # Get individual instances