diff --git a/src/aspire/basis/fspca.py b/src/aspire/basis/fspca.py
index 16500918af..c627a64e94 100644
--- a/src/aspire/basis/fspca.py
+++ b/src/aspire/basis/fspca.py
@@ -4,7 +4,6 @@
 import numpy as np
 
 from aspire.basis import Coef, ComplexCoef, FFBBasis2D, SteerableBasis2D
-from aspire.covariance import BatchedRotCov2D
 from aspire.operators import BlkDiagMatrix
 from aspire.utils import complex_type, fix_signs, real_type
 
@@ -147,6 +146,9 @@ def build(self):
             self.noise_var = WhiteNoiseEstimator(self.src).estimate()
         logger.info(f"Setting noise_var={self.noise_var}")
 
+        # Import BatchedRotCov2D here to prevent circular imports.
+        from aspire.covariance import BatchedRotCov2D
+
         cov2d = BatchedRotCov2D(
             src=self.src, basis=self.basis, batch_size=self.batch_size
         )
diff --git a/src/aspire/utils/resolution_estimation.py b/src/aspire/utils/resolution_estimation.py
index 4f97c22e7d..14c142bdc5 100644
--- a/src/aspire/utils/resolution_estimation.py
+++ b/src/aspire/utils/resolution_estimation.py
@@ -8,8 +8,7 @@
 import matplotlib.pyplot as plt
 import numpy as np
 
-from aspire.nufft import nufft
-from aspire.numeric import fft
+from aspire import nufft, numeric
 from aspire.utils import grid_2d, grid_3d
 
 logger = logging.getLogger(__name__)
@@ -152,8 +151,8 @@ def _fft_correlations(self):
         ]
 
         # Compute centered Fourier transforms.
-        f1 = fft.centered_fftn(self.a, axes=self._fourier_axes)
-        f2 = fft.centered_fftn(self.b, axes=self._fourier_axes)
+        f1 = numeric.fft.centered_fftn(self.a, axes=self._fourier_axes)
+        f2 = numeric.fft.centered_fftn(self.b, axes=self._fourier_axes)
 
         # Construct an output table of correlations
         correlations = np.zeros(
@@ -232,7 +231,7 @@ def _nufft_correlations(self):
         #   Note, we want a complex result.
         signal = np.vstack((self._a, self._b))
         # Compute one large NUFFT for all the signal frames,
-        f = nufft(signal, fourier_pts, real=False)
+        f = nufft.nufft(signal, fourier_pts, real=False)
         # then unpack as two 1D stacks of the polar grid points, one for _a and _b.
         f = f.reshape(self._a.shape[0] + self._b.shape[0], len(r), -1)
         f1, f2 = np.vsplit(f, [self._a.shape[0]])