feat: add best sparse model flag

Add BestSparseModel utility to detect and cache the sparse model
with the most registered images, allowing intrinsics refinement and
transforms.json generation to reliably use the best reconstruction.

Author: luishresende

nerfstudio/process_data/colmap_converter_to_nerfstudio_dataset.py

@@ -44,6 +44,9 @@ class ColmapConverterToNerfstudioDataset(BaseConverterToNerfstudioDataset):
     refine_intrinsics: bool = True
     """If True, do bundle adjustment to refine intrinsics.
     Only works with colmap sfm_tool"""
+    use_best_sparse_model: bool = True
+    """If True, use the best sparse model to refine intrinsics and save the camera transformations.
+    Only works with colmap sfm_tool"""
     feature_type: Literal[
         "any",
         "sift",
@@ -111,6 +114,8 @@ def default_colmap_path() -> Path:
 
     @property
     def absolute_colmap_model_path(self) -> Path:
+        if self.use_best_sparse_model:
+            return colmap_utils.BestSparseModel.get_model_path(self.absolute_colmap_path)
         return self.output_dir / self.colmap_model_path
 
     @property
@@ -218,6 +223,7 @@ def _run_colmap(self, mask_path: Optional[Path] = None):
                 verbose=self.verbose,
                 matching_method=self.matching_method,
                 refine_intrinsics=self.refine_intrinsics,
+                use_best_sparse_model=self.use_best_sparse_model,
                 colmap_cmd=self.colmap_cmd,
             )
         elif sfm_tool == "hloc":

nerfstudio/process_data/colmap_utils.py

@@ -98,6 +98,7 @@ def run_colmap(
     verbose: bool = False,
     matching_method: Literal["vocab_tree", "exhaustive", "sequential"] = "vocab_tree",
     refine_intrinsics: bool = True,
+    use_best_sparse_model: bool = True,
     colmap_cmd: str = "colmap",
 ) -> None:
     """Runs COLMAP on the images.
@@ -111,6 +112,7 @@ def run_colmap(
         verbose: If True, logs the output of the command.
         matching_method: Matching method to use.
         refine_intrinsics: If True, refine intrinsics.
+        use_best_sparse_model: If True, refine intrinsics of the best sparse model.
         colmap_cmd: Path to the COLMAP executable.
     """
 
@@ -173,11 +175,13 @@ def run_colmap(
     CONSOLE.log("[bold green]:tada: Done COLMAP bundle adjustment.")
 
     if refine_intrinsics:
+        sparse_model = "0" if not use_best_sparse_model else BestSparseModel.get_model(colmap_dir)
+
         with status(msg="[bold yellow]Refine intrinsics...", spinner="dqpb", verbose=verbose):
             bundle_adjuster_cmd = [
                 f"{colmap_cmd} bundle_adjuster",
-                f"--input_path {sparse_dir}/0",
-                f"--output_path {sparse_dir}/0",
+                f"--input_path {sparse_dir}/{sparse_model}",
+                f"--output_path {sparse_dir}/{sparse_model}",
                 "--BundleAdjustment.refine_principal_point 1",
             ]
             run_command(" ".join(bundle_adjuster_cmd), verbose=verbose)
@@ -712,3 +716,64 @@ def create_ply_from_colmap(
             x, y, z = coord
             r, g, b = color
             f.write(f"{x:8f} {y:8f} {z:8f} {r} {g} {b}\n")
+
+
+class BestSparseModel:
+    """
+    Utility class to find and cache the best COLMAP sparse model
+    from a given directory. Provides both the model name and full path.
+
+    The best model is defined as the one with the largest number of registered images.
+    """
+
+    _cached_model = None  # Cached name of the best sparse model
+
+    @staticmethod
+    def _find_best_model(colmap_dir: Path) -> str:
+        """
+        Find the best sparse model in the COLMAP directory.
+
+        Args:
+            colmap_dir (Path): Path to the COLMAP project directory containing 'sparse/'.
+
+        Returns:
+            str: Name of the best sparse model directory.
+
+        Raises:
+            ValueError: If no valid sparse models are found.
+        """
+        sparse_dir = colmap_dir / "sparse"
+        models = [m for m in sorted(sparse_dir.glob("*")) if (m / "images.bin").exists()]
+        if not models:
+            raise ValueError(f"No valid COLMAP sparse models found in {sparse_dir}")
+
+        best_model = max(models, key=lambda m: len(read_images_binary(m / "images.bin")))
+        return best_model.name
+
+    @staticmethod
+    def get_model(colmap_dir: Path) -> str:
+        """
+        Get the name of the best sparse model, using cache if available.
+
+        Args:
+            colmap_dir (Path): Path to the COLMAP project directory containing 'sparse/'.
+
+        Returns:
+            str: Name of the best sparse model.
+        """
+        if not BestSparseModel._cached_model:
+            BestSparseModel._cached_model = BestSparseModel._find_best_model(colmap_dir)
+        return BestSparseModel._cached_model
+
+    @staticmethod
+    def get_model_path(colmap_dir: Path) -> Path:
+        """
+        Get the full path to the best sparse model, using cache if available.
+
+        Args:
+            colmap_dir (Path): Path to the COLMAP project directory containing 'sparse/'.
+
+        Returns:
+            Path: Full path to the best sparse model directory.
+        """
+        return colmap_dir / "sparse" / BestSparseModel.get_model(colmap_dir)