From 3cd48f7e5faddbd21d187336df0b27fb12668ee0 Mon Sep 17 00:00:00 2001
From: Stuart Mumford <stuart@cadair.com>
Date: Wed, 6 Mar 2024 16:00:55 +0000
Subject: [PATCH 1/2] First pass at a vbi_reproject function

---
 dkist/processing/__init__.py   |   0
 dkist/processing/mosaicking.py | 118 +++++++++++++++++++++++++++++++++
 2 files changed, 118 insertions(+)
 create mode 100644 dkist/processing/__init__.py
 create mode 100644 dkist/processing/mosaicking.py

diff --git a/dkist/processing/__init__.py b/dkist/processing/__init__.py
new file mode 100644
index 000000000..e69de29bb
diff --git a/dkist/processing/mosaicking.py b/dkist/processing/mosaicking.py
new file mode 100644
index 000000000..e76170579
--- /dev/null
+++ b/dkist/processing/mosaicking.py
@@ -0,0 +1,118 @@
+"""
+Helper functions for stitching (VBI) frames.
+"""
+
+import copy
+from typing import Callable
+
+import numpy as np
+import reproject
+import reproject.mosaicking
+import tqdm
+
+import astropy.units as u
+from astropy.table import Table
+from gwcs import WCS
+from gwcs import coordinate_frames as cf
+
+import dkist
+
+
+def _unify_output_frames(tds: dkist.TiledDataset, reference_tile: tuple[slice]):
+    """
+    Given a `dkist.TiledDataset` return a new version where all the WCSes share
+    the same celestial output frame.
+    """
+    ref_wcs = tds[reference_tile].wcs
+    ref_celestial_frame = [f for f in ref_wcs.output_frame.frames if isinstance(f, cf.CelestialFrame)][0]
+    new_datasets = []
+    for ds in tds.flat:
+        celestial_ind = int(np.where([isinstance(f, cf.CelestialFrame) for f in ds.wcs.output_frame.frames])[0][0])
+        new_frames = copy.deepcopy(ds.wcs.output_frame.frames)
+        new_frames[celestial_ind] = ref_celestial_frame
+        new_wcs = WCS(
+            ds.wcs.forward_transform,
+            input_frame=ds.wcs.input_frame,
+            output_frame=cf.CompositeFrame(new_frames),
+        )
+        new_ds = dkist.Dataset(
+            ds.data,
+            new_wcs,
+            meta=ds.meta,
+            unit=ds.unit,
+            mask=ds.mask,
+        )
+        new_ds._file_manager = ds._file_manager
+        new_datasets.append(new_ds)
+
+    return dkist.TiledDataset(
+        np.array(new_datasets).reshape(tds.shape),
+        inventory=tds.inventory,
+    )
+
+
+def reproject_vbi(
+    tds: dkist.TiledDataset,
+    *,
+    edge_crop: int = 50,
+    reference_tile: tuple[slice] = np.s_[2, 0],
+    reproject_function: Callable,
+    roundtrip_coords: bool = False,
+    combine_function: str = "first",
+    shape_out: tuple[int] = None,
+):
+    uni_tds = _unify_output_frames(tds, reference_tile)
+    cropped = dkist.TiledDataset(
+        np.array([ds[:, edge_crop:-edge_crop, edge_crop:-edge_crop] for ds in uni_tds.flat]).reshape(tds.shape),
+        inventory=tds.inventory,
+    )
+
+    target_shape = shape_out or np.array(cropped[reference_tile][0].data.shape) * cropped.shape
+
+    # We are going to use the reference_tile's WCS to create our output WCS
+    ref_tile = tds[reference_tile]
+    # Get the model for the celestial coords of the first image in the ref tile
+    celestial = ref_tile.wcs.forward_transform[0].transform_at_index(0)
+    # We are using the timesteps for the ref tile
+    # TODO: use the mean time of all the tiles and align this with obstime in the celestial frame
+    temporal = ref_tile.wcs.forward_transform[1]
+
+    target_celestial_wcs = WCS(
+        forward_transform=celestial,
+        input_frame=cf.CoordinateFrame(2, ["PIXEL", "PIXEL"], (0, 1), unit=(u.pix, u.pix)),
+        output_frame=ref_tile.wcs.output_frame.frames[0],
+    )
+
+    target_full_wcs = WCS(
+        forward_transform=celestial & temporal,
+        input_frame=cf.CoordinateFrame(3, ["PIXEL"]*3, (0, 1, 2), unit=[u.pix]*3),
+        output_frame=ref_tile.wcs.output_frame,
+    )
+
+    output = []
+    footprint = []
+    for tind in tqdm.tqdm(range(uni_tds.flat[0].data.shape[0])):
+        tiles = [ds[tind] for ds in cropped.flat]
+        arr, fp = reproject.mosaicking.reproject_and_coadd(
+            tiles,
+            target_celestial_wcs,
+            shape_out=target_shape,
+            reproject_function=reproject_function,
+            roundtrip_coords=roundtrip_coords,
+            combine_function=combine_function,
+        )
+        output.append(arr)
+        footprint.append(fp)
+
+    output = np.array(output)
+    footprint = np.array(footprint)
+
+    new_ds = dkist.Dataset(
+        output,
+        wcs=target_full_wcs,
+        unit=ref_tile.unit,
+        # TODO: hmm?
+        meta={"headers": Table(), "inventory": tds.inventory},
+    )
+
+    return new_ds, footprint

From 6512393ad9a32bccaee5989c46043d617f3c2a06 Mon Sep 17 00:00:00 2001
From: Stuart Mumford <stuart@cadair.com>
Date: Wed, 6 Mar 2024 17:10:37 +0000
Subject: [PATCH 2/2] Implement combine by r0

---
 dkist/processing/mosaicking.py | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/dkist/processing/mosaicking.py b/dkist/processing/mosaicking.py
index e76170579..f456a7768 100644
--- a/dkist/processing/mosaicking.py
+++ b/dkist/processing/mosaicking.py
@@ -93,6 +93,12 @@ def reproject_vbi(
     footprint = []
     for tind in tqdm.tqdm(range(uni_tds.flat[0].data.shape[0])):
         tiles = [ds[tind] for ds in cropped.flat]
+        if combine_function == "fried_parameter":
+            # Sort the tiles by the fried parameter so the first one is
+            # prioritised with the best seeing.
+            tiles = list(sorted(tiles, key=lambda d: d.headers[0]["ATMOS_R0"]))
+            combine_function = "first"
+
         arr, fp = reproject.mosaicking.reproject_and_coadd(
             tiles,
             target_celestial_wcs,