Introduce HoloDataset class to include loading data from .h5 and .tif files for holography reconstructions

src/cdtools/datasets/holo_dataset.py

@@ -0,0 +1,79 @@
+from torch.utils.data import Dataset as torchDataset
+import torch as t
+import numpy as np
+import h5py
+import pathlib
+from PIL import Image
+
+
+class HoloDataset(torchDataset):
+    def __init__(self, data, transform=None):
+        self.data = data
+        self.transform = transform
+
+    @classmethod
+    def from_h5(cls, h5_file, path_within_h5, axes_to_average=None, device=None, replace_dead_pixels=True):
+        """Generates a new HoloDataset from a .h5 file directly
+
+        Parameters
+        ----------
+        h5_file : str, pathlib.Path, or h5py.File
+            The .h5 file to load from
+        path_within_h5 : str
+            The path within the h5 file to load the data from
+        axes_to_average : tuple, optional
+            The axes to average the data over
+        device : torch.device, optional
+            The device to load the data onto
+
+        Returns
+        -------
+        dataset : HoloDataset
+            The constructed dataset object
+        """
+        
+        # If a bare string is passed
+        if isinstance(h5_file, str) or isinstance(h5_file, pathlib.Path):
+            with h5py.File(h5_file,'r') as f:
+                return cls.from_h5(f, path_within_h5, axes_to_average, device)
+        
+        data = t.tensor(h5_file[path_within_h5][:,:,:,:], device=device, dtype=t.float32)
+        if axes_to_average is not None:
+            data = t.mean(data, dim=axes_to_average)
+
+        # Replace dead pixels with the mean of the surrounding pixels
+        if replace_dead_pixels:
+            dead_pixels = list(t.argwhere(data == 0))
+            for n in dead_pixels:
+                data[n[0],n[1]] = t.mean(t.tensor([data[n[0]+1,n[1]+1], data[n[0]+1,n[1]], data[n[0],n[1]+1], data[n[0]-1,n[1]-1], data[n[0]-1,n[1]], data[n[0],n[1]-1]]))
+        
+        return cls(data)
+
+    @classmethod
+    def from_tif(cls, tif_file):
+        """Generates a new HoloDataset from a .tif file directly. It assumes that the .tif file is 
+        already stitched, edited and is a single pattern.
+
+        Parameters
+        ----------
+        tif_file : str, pathlib.Path
+            The .tif file to load from
+
+        Returns
+        -------
+        dataset : HoloDataset
+            The constructed dataset object
+        """
+        data = t.tensor(np.array(Image.open(tif_file)), dtype=t.float32)
+        return cls(data)
+
+    def __len__(self):
+        return 1  # Since we're only loading one pattern
+
+    def __getitem__(self, idx):
+        data = self.data
+        if self.transform:
+            data = self.transform(data)
+        return data
+    
+