diff --git a/.gitignore b/.gitignore
index 68787a8177f51b86162e68d12991aa189d5e5983..a5bc312c13cbc0b20920de4c9e80d5ba31f77a2a 100644
--- a/.gitignore
+++ b/.gitignore
@@ -160,4 +160,6 @@ cython_debug/
 #.idea/
 
 
-outputs/
\ No newline at end of file
+outputs/
+
+fileForMessingAround.py
\ No newline at end of file
diff --git a/PyNutil/PyNutil.py b/PyNutil/PyNutil.py
index d0df0b9c7c04de51d13ba9a2873559d0803d5bff..42853a0e64948012e4fdbc95b4af69ad222861da 100644
--- a/PyNutil/PyNutil.py
+++ b/PyNutil/PyNutil.py
@@ -2,7 +2,7 @@
 import numpy as np
 from DeepSlice.coord_post_processing.spacing_and_indexing import number_sections
 import json
-from VisuAlignWarpVec import triangulate, transform_vec
+from visualign_deformations import triangulate, transform_vec
 from glob import glob
 from tqdm import tqdm
 import cv2
diff --git a/PyNutil/folder_of_segmentations_to_meshview.py b/PyNutil/folder_of_segmentations_to_meshview.py
new file mode 100644
index 0000000000000000000000000000000000000000..6505920a5c4df96e5aa7ea3922794457ddbd97e0
--- /dev/null
+++ b/PyNutil/folder_of_segmentations_to_meshview.py
@@ -0,0 +1,84 @@
+
+#pandas is used for working with csv files
+import pandas as pd
+#nrrd just lets us open nrrd files
+import nrrd
+import numpy as np
+import csv
+
+from datetime import datetime
+
+#import our function for converting a folder of segmentations to points
+from PyNutil import FolderToAtlasSpace, labelPoints, WritePointsToMeshview
+
+volume_path = "../annotation_volumes//annotation_10_reoriented.nrrd"
+data, header = nrrd.read(volume_path)
+
+startTime = datetime.now()
+
+segmentation_folder = "../test_data/oneSection15/"
+alignment_json = "../test_data/C68_nonlinear.json"
+#now we can use our function to convert the folder of segmentations to points
+points = FolderToAtlasSpace(segmentation_folder,alignment_json, pixelID=[255, 0, 255], nonLinear=True)
+#first we need to find the label file for the volume
+label_path = "../annotation_volumes//allen2022_colours.csv"
+#then the path to the volume
+
+#read the label files
+label_df = pd.read_csv(label_path)
+#read the annotation volume, it also has a header but we don't need it
+#now we can get the labels for each point
+labels = labelPoints(points, data, scale_factor=2.5)
+#save points to a meshview json
+WritePointsToMeshview(points, labels,"../outputs/points.json", label_df)
+
+#Task:
+# Make a pandas dataframe
+# Column 1: counted_labels
+# Column 2: label_counts
+# Column 3: region_name (look up by reading Allen2022_colours.csv, look up name and RGB)
+# Save dataframe in output as CSV
+# next task is to create functions from this. 
+counted_labels, label_counts = np.unique(labels, return_counts=True)
+counts_per_label = list(zip(counted_labels,label_counts))
+
+df_counts_per_label = pd.DataFrame(counts_per_label, columns=["allenID","pixel count"])
+
+allen_colours = "../annotation_volumes//allen2022_colours.csv"
+
+df_allen_colours =pd.read_csv(allen_colours, sep=",")
+df_allen_colours
+
+#look up name, r, g, b in df_allen_colours in df_counts_per_label based on "allenID"
+new_rows = []
+for index, row in df_counts_per_label.iterrows():
+    mask = df_allen_colours["allenID"] == row["allenID"] 
+    current_region_row = df_allen_colours[mask]
+    current_region_name = current_region_row["name"].values
+    current_region_red = current_region_row["r"].values
+    current_region_green = current_region_row["g"].values
+    current_region_blue = current_region_row["b"].values
+
+    row["name"]  = current_region_name[0]
+    row["r"] = current_region_red[0]
+    row["g"] = current_region_green[0]
+    row["b"] = current_region_blue[0]
+    
+    new_rows.append(row)
+
+df_counts_per_label_name = pd.DataFrame(new_rows)
+df_counts_per_label_name
+
+# write to csv file
+df_counts_per_label_name.to_csv("../outputs/counts_per_allenID.csv", sep=";", na_rep='', index= False)
+
+#r = df_allen_colours["r"]
+#g = df_allen_colours["g"]
+#b = df_allen_colours["b"]
+#region_name = df_allen_colours["name"]
+
+#while we havent added it here it would be good to next quantify the number of cells for each label.
+
+time_taken = datetime.now() - startTime
+
+print(f"time taken was: {time_taken}")
\ No newline at end of file
diff --git a/workflows/folder_of_segmentations_to_meshview.py b/workflows/folder_of_segmentations_to_meshview.py
deleted file mode 100644
index d90da81591fe38eb281b091cd06d0f0dc65be3a7..0000000000000000000000000000000000000000
--- a/workflows/folder_of_segmentations_to_meshview.py
+++ /dev/null
@@ -1,31 +0,0 @@
-#pandas is used for working with csv files
-import pandas as pd
-#nrrd just lets us open nrrd files
-import nrrd
-
-#this is a hack to let us import as if we are in the parent directory
-import sys
-sys.path.append('..')
-#import our function for converting a folder of segmentations to points
-from PyNutil.PyNutil import FolderToAtlasSpace, labelPoints, WritePointsToMeshview
-
-
-segmentation_folder = "../test_data/oneSection15/"
-alignment_json = "../test_data/C68_nonlinear.json"
-#now we can use our function to convert the folder of segmentations to points
-points = FolderToAtlasSpace(segmentation_folder,alignment_json, pixelID=[255, 0, 255], nonLinear=True)
-#first we need to find the label file for the volume
-label_path = "../annotation_volumes//allen2022_colours.csv"
-#then the path to the volume
-volume_path = "../annotation_volumes//annotation_10_reoriented.nrrd"
-#read the label file
-label_df = pd.read_csv(label_path)
-#read the annotation volume, it also has a header but we don't need it
-data, header = nrrd.read(volume_path)
-#now we can get the labels for each point
-labels = labelPoints(points, data, scale_factor=2.5)
-#save points to a meshview json
-WritePointsToMeshview(points, labels,"../outputs/points.json", label_df)
-
-#while we havent added it here it would be good to next quantify the number of cells for each label.
-