Merge pull request #7 from vectoraico/Shehryar718-patch-1

Python COLMAP

python_colmap.py

@@ -0,0 +1,90 @@
+import pathlib
+import pycolmap
+import numpy as np
+import cv2
+import matplotlib.pyplot as plt
+from IPython.display import display, Image
+
+# Function to visualize images
+def display_images(image_paths):
+    for path in image_paths:
+        display(Image(filename=str(path)))
+
+# Function to visualize SIFT features
+def visualize_sift_features(database_path, num_images=3):
+    # Load the database
+    database = pycolmap.read_database(database_path)
+
+    # Randomly select images for visualization
+    selected_image_ids = np.random.choice(list(database['images'].keys()), num_images, replace=False)
+
+    for image_id in selected_image_ids:
+        image = database['images'][image_id]
+        keypoints = image['point2D']
+
+        img = cv2.imread(str(image['name']))
+
+        # Draw SIFT features on the image
+        img_with_sift = cv2.drawKeypoints(
+            img,
+            [cv2.KeyPoint(x=k[0], y=k[1], _size=k[2], _angle=k[3], _response=k[4], _octave=k[5], _class_id=k[6]) for k in keypoints],
+            None,
+            flags=cv2.DRAW_MATCHES_FLAGS_DRAW_RICH_KEYPOINTS
+        )
+
+        plt.imshow(cv2.cvtColor(img_with_sift, cv2.COLOR_BGR2RGB))
+        plt.title(f"SIFT Features - Image {image_id}")
+        plt.axis('off')
+        plt.show()
+
+# Function to visualize 3D reconstruction
+def visualize_3d_reconstruction(reconstruction_path):
+    colmap_path = pathlib.Path("./mvs")
+    !{colmap_path / 'gui'} {reconstruction_path}
+
+# Function to reconstruct with options and visualize at each step
+def reconstruct_with_options_and_visualize(image_folder, output_path):
+    # Extract features
+    database_path = output_path / "database.db"
+    pycolmap.extract_features(
+        database_path,
+        image_folder,
+        sift_options={"max_num_features": 10000, "first_octave": -1, "num_octaves": 5}
+    )
+
+    # Visualize extracted features on random images
+    visualize_sift_features(database_path)
+
+    # Match features
+    pycolmap.match_exhaustive(
+        database_path,
+        matching_options={"SiftMatchingOptions": {"max_num_matches": 32768, "lowes_ratio": 0.8}}
+    )
+
+    # Incremental reconstruction
+    maps = pycolmap.incremental_mapping(
+        database_path,
+        image_folder,
+        output_path,
+        mapper_options={"MapperOptions": {"extract_colors": True}}
+    )
+
+    # Visualize sparse reconstruction
+    visualize_3d_reconstruction(output_path / "sparse/0")
+
+    # Dense reconstruction
+    mvs_path = output_path / "mvs"
+    pycolmap.undistort_images(mvs_path, output_path, image_folder)
+
+    # Optional: Increase max_gcp_distance for better dense reconstruction results
+    # Uses CUDA
+    pycolmap.patch_match_stereo(mvs_path, {"PatchMatchStereoOptions": {"max_gcp_distance": 60}})
+    pycolmap.stereo_fusion(mvs_path / "dense.ply", mvs_path)
+
+    # Visualize dense reconstruction (optional)
+    visualize_3d_reconstruction(mvs_path / "dense.ply")
+
+image_folder = pathlib.Path("./image_folder")
+output_path = pathlib.Path("./output_path")
+
+reconstruct_with_options_and_visualize(image_folder, output_path)