diff --git a/src/nyx/features/focus_score.cpp b/src/nyx/features/focus_score.cpp
index b5c0df18..43a62212 100644
--- a/src/nyx/features/focus_score.cpp
+++ b/src/nyx/features/focus_score.cpp
@@ -3,6 +3,10 @@
 
 using namespace Nyxus;
 
+int FocusScoreFeature::kernel[9] = { 0, 1, 0, 
+                                     1, -4, 1, 
+                                     0, 1, 0 };
+
 FocusScoreFeature::FocusScoreFeature() : FeatureMethod("FocusScoreFeature") {
     provide_features(FocusScoreFeature::featureset);
 }
@@ -70,6 +74,19 @@ void FocusScoreFeature::reduce (size_t start, size_t end, std::vector<int>* ptrL
     }
 }
 
+void FocusScoreFeature::osized_calculate(LR& r, ImageLoader& imloader) {
+
+    // Skip calculation in case of noninformative data
+    if (r.aux_max == r.aux_min) return;
+
+    WriteImageMatrix_nontriv Im0 ("FocusScoreFeature-osized_calculate-Im0", r.label);
+    Im0.allocate_from_cloud (r.raw_pixels_NT, r.aabb, false);
+
+    focus_score_ = get_focus_score_NT(Im0, 1);
+
+    save_value(r.fvals);
+}
+
 void FocusScoreFeature::save_value(std::vector<std::vector<double>>& feature_vals) {
     
     feature_vals[(int)FeatureIMQ::FOCUS_SCORE][0] = focus_score_;
@@ -77,6 +94,102 @@ void FocusScoreFeature::save_value(std::vector<std::vector<double>>& feature_val
 
 }
 
+double FocusScoreFeature::get_focus_score_NT(WriteImageMatrix_nontriv& Im, int ksize) {
+
+    int n = 3; // size of kernel nxn
+
+    if (ksize != 1) {
+
+        kernel[0] = 2; 
+        kernel[1] = 0;
+        kernel[2] = 2;
+
+        kernel[3] = 0;
+        kernel[4] = -8;
+        kernel[5] = 0;
+
+        kernel[6] = 2;
+        kernel[7] = 0;
+        kernel[8] = 2;
+    }
+
+    auto width = Im.get_width(),
+        height = Im.get_height(); 
+    auto xy0 = (int)ceil(double(n) / 2.);
+
+    // Window (N-fold kernel size)
+    int winX = n * 10, 
+        winY = n * 10;
+    std::vector<unsigned int> W (winY * winX);
+
+    // Convolution result buffer
+    std::vector<double> conv_buffer ((winY + n - 1) * (winX + n - 1) * 2);
+
+    // Iterate the image window by window
+    int n_winHor = width / winX, //ceil (float(width) / float(win)), 
+        n_winVert = height / winY; //ceil (float(height) / float(win));
+
+    
+    // ROI smaller than kernel?
+    if (n_winVert == 0 || n_winHor == 0)
+    {
+        // Fill the window with data
+        for (int row=0; row < height; row++) {
+            for (int col = 0; col < width; col++)
+            {
+                size_t idx = row * width + col;
+                W[idx] = Im.get_at(idx);
+            }    
+        }
+
+        // Convolve
+        laplacian (W, conv_buffer, width, height, ksize);
+        
+        return variance(conv_buffer);
+    }
+
+    // Variables for Welford Algorithm for calculating variance
+    double mean = 0.;
+    double M2 = 0.; // sum of squared distance
+    int count = 0;
+
+    std::vector<std::tuple<double, double, int>> tile_variance; // vector of tuples containing 0: abs sum of tile, 1: variance of tile, 2: buffer size
+
+    // ROI larger than kernel
+    // Calculate laplacian of window and then use Welford algorithm to calculate variance of image
+    for (int winVert = 0; winVert < n_winVert; winVert++)
+    {
+        for (int winHor = 0; winHor < n_winHor; winHor++)
+        {
+
+            // Fill the window with data
+            for (int row=0; row<winY; row++) {
+                for (int col = 0; col < winX; col++)
+                {
+                    size_t imIdx = winVert * n_winHor * winY * winX // skip whole windows above
+                        + row * n_winHor * winX  // skip 'row' image-wide horizontal lines
+                        + winHor * n_winHor * winX   // skip winHor-wide line
+                        + col;
+                    W[row * winX + col] = Im.get_at(imIdx);
+                }
+            }
+
+            // Convolve
+            laplacian (W, conv_buffer, winY, winX, ksize);
+
+            // Calculate count, mean, and sum of squared differences (M2)
+            for (const auto& pixel: conv_buffer) {
+                ++count;
+                double delta = pixel - mean;
+                mean += delta / count;
+                M2 += delta * (pixel - mean);
+            }
+        }
+    }
+
+    return M2 / count; // Return variance
+}
+
 double FocusScoreFeature::get_local_focus_score(const std::vector<PixIntens>& image, int height, int width, int ksize, int scale) {
 
     double local_focus_score = 0;
@@ -113,18 +226,15 @@ void FocusScoreFeature::laplacian(const std::vector<PixIntens>& image, std::vect
     int m_kernel = 3;
     int n_kernel = 3;
 
-    // use c-style array for compile time initialization
-    int kernel[9] = { 0, 1, 0, 
-                    1, -4, 1, 
-                    0, 1, 0 };
-
     if (ksize != 1) {
         kernel[0] = 2; 
         kernel[1] = 0;
         kernel[2] = 2;
+
         kernel[3] = 0;
         kernel[4] = -8;
         kernel[5] = 0;
+
         kernel[6] = 2;
         kernel[7] = 0;
         kernel[8] = 2;
diff --git a/src/nyx/features/focus_score.h b/src/nyx/features/focus_score.h
index 7119f7c4..69d6c3ee 100644
--- a/src/nyx/features/focus_score.h
+++ b/src/nyx/features/focus_score.h
@@ -26,7 +26,8 @@ class FocusScoreFeature: public FeatureMethod
 
     //=== Non-trivial ROIs ===
     void osized_add_online_pixel(size_t x, size_t y, uint32_t intensity) {}
-    void osized_calculate(LR& r, ImageLoader& imloader){};
+    void osized_calculate(LR& r, ImageLoader& imloader);
+    double get_focus_score_NT(WriteImageMatrix_nontriv& Im, int ksize);
 
     // Result saver
     void save_value(std::vector<std::vector<double>>& feature_vals);
@@ -50,5 +51,6 @@ class FocusScoreFeature: public FeatureMethod
 
     static double variance(const std::vector<double>& image);
 
+    static int kernel[9];
 };
 
diff --git a/src/nyx/python/nyxus/nyxus.py b/src/nyx/python/nyxus/nyxus.py
index 24bd5c2f..1ce1efea 100644
--- a/src/nyx/python/nyxus/nyxus.py
+++ b/src/nyx/python/nyxus/nyxus.py
@@ -1296,7 +1296,7 @@ def __init__(
             'neighbor_distance', 'pixels_per_micron', 'coarse_gray_depth',
             'n_feature_calc_threads', 'n_loader_threads', 'ibsi',
             'gabor_kersize', 'gabor_gamma', 'gabor_sig2lam', 'gabor_f0',
-            'channel_signature', 'min_intensity', 'max_intensity'
+            'channel_signature', 'min_intensity', 'max_intensity', 'ram_limit',
         }
 
         # Check for unexpected keyword arguments
@@ -1316,6 +1316,7 @@ def __init__(
         dynamic_range = kwargs.get('dynamic_range', 10000)
         min_intensity = kwargs.get('min_intensity', 0.0)
         max_intensity = kwargs.get('max_intensity', 1.0)
+        ram_limit = kwargs.get('ram_limit', -1)
         
         if neighbor_distance <= 0:
             raise ValueError("Neighbor distance must be greater than zero.")
@@ -1339,8 +1340,6 @@ def __init__(
         if(using_gpu > -1 and not gpu_available()):
             print("No gpu available.")
             using_gpu = -1
-    
-        ram_limit = -1
 
         initialize_environment(
             2, # 2D