Update terramesh.py

terramesh.py

@@ -19,11 +19,8 @@
 import os
 import io
 import re
-
-import numpy
 import zarr
 import fsspec
-import itertools
 import braceexpand
 import numpy as np
 import albumentations
@@ -52,10 +49,11 @@ split_files = {
 
 def build_terramesh_dataset(
         path: str = "https://huggingface.co/datasets/ibm-esa-geospatial/TerraMesh/resolve/main/",
-        modalities=None,
+        modalities: list | str = None,
         split: str = "val",
         urls: str | None = None,
         batch_size: int = 8,
+        return_metadata: bool = False,
         *args, **kwargs,
 ):
     if len(modalities) == 1:
@@ -66,6 +64,7 @@ def build_terramesh_dataset(
             split=split,
             urls=urls,
             batch_size=batch_size,
+            return_metadata=return_metadata,
             *args, **kwargs
         )
         return dataset
@@ -78,6 +77,7 @@ def build_terramesh_dataset(
             split=split,
             urls=urls,
             batch_size=batch_size,
+            return_metadata=return_metadata,
             *args, **kwargs,
         )
         return dataset
@@ -86,7 +86,28 @@ def build_terramesh_dataset(
 def zarr_decoder(key, value):
     if key == "zarr.zip" or key.endswith(".zarr.zip"):
         mapper = fsspec.filesystem("zip", fo=io.BytesIO(value), block_size=None).get_mapper("")
-        return zarr.open_consolidated(mapper, mode="r")['bands'][...]
+        return zarr.open_consolidated(mapper, mode="r")["bands"][...]
+
+
+def zarr_metadata_decoder(sample):
+    for key, value in list(sample.items()):
+        if key == "zarr.zip" or key.endswith(".zarr.zip"):
+            mapper = fsspec.filesystem("zip", fo=io.BytesIO(value), block_size=None).get_mapper("")
+            data = zarr.open_consolidated(mapper, mode="r")
+            sample[key] = data["bands"][...]
+
+            # Add metadata
+            if "center_lon" not in sample.keys():  # Same center point for all modalities
+                sample["center_lon"] = data["center_lon"][...]
+                sample["center_lat"] = data["center_lat"][...]
+            if "cloud_mask" in data and "cloud_mask" not in sample.keys():  # Same S2 mask in all optical modalities
+                sample["cloud_mask"] = data["cloud_mask"][...][np.newaxis, ...]  # Add channel dim to mask
+            if data["time"][...] > 1e6:  # DEM has no valid timestamp (value = 0)
+                time_key = "time" if key == "zarr.zip" else "time_" + key
+                sample[time_key] = data["time"][...]  # Integer values of type "datetime64[ns]"
+        # TODO Other types are currently not decoded, fall back to autodecode
+
+    return sample
 
 
 def identity(sample):
@@ -115,6 +136,7 @@ def build_wds_dataset(
         urls: str | None = None,
         batch_size: int = 8,
         transform: Callable = None,
+        return_metadata: bool = False,
         *args, **kwargs
 ):
     if urls is None:
@@ -134,12 +156,16 @@ def build_wds_dataset(
     kwargs["shardshuffle"] = kwargs.get("shardshuffle", 100)  # Shuffle shard by default
 
     # Build dataset
-    dataset = (
-        wds.WebDataset(urls, *args, **kwargs)
-        .decode(zarr_decoder)  # Decode byte files
-        .rename(image='zarr.zip')
-        .map_dict(image=drop_time_dim)  # Remove temporal dimension
-    )
+    dataset = wds.WebDataset(urls, *args, **kwargs)
+
+    # Decode from bytes to numpy arrays, etc.
+    dataset = dataset.map(zarr_metadata_decoder) if return_metadata else dataset.decode(zarr_decoder)
+
+    # Rename modality to "image" and remove temporal dimension
+    dataset = (dataset
+               .rename(image="zarr.zip")
+               .map(drop_time_dim)
+               )
 
     if transform is not None:
         dataset = dataset.map(transform)
@@ -151,19 +177,18 @@ def build_wds_dataset(
     return dataset
 
 
-def combine_datasets(*args):
-    return itertools.chain(*args)
-
-
 def build_multimodal_dataset(
         path: str = "https://huggingface.co/datasets/ibm-esa-geospatial/TerraMesh/resolve/main/",
-        modalities: str = "S2L2A",
+        modalities: list = None,
         split: str = "val",
         urls: str | None = None,
         batch_size: int = 8,
         transform: Callable = None,
+        return_metadata: bool = False,
         *args, **kwargs
 ):
+    if modalities is None:
+        modalities = ["S2L2A", "S2L1C", "S2RGB", "S1GRD", "S1RTC", "DEM", "NDVI", "LULC"]  # Default
     if urls is None:
         # Filter modalities based availability (S1GRD and S1RTC not present in all subsets)
         def filter_list(lst, value):
@@ -180,17 +205,21 @@ def build_multimodal_dataset(
         urls = (os.path.join(path, split, majortom_mod, split_files["majortom"][split][0])
                 + "::" + os.path.join(path, split, ssl4eos12_mod, split_files["ssl4eos12"][split][0]))
 
-    dataset = build_datapipeline(urls, transform, batch_size, *args, **kwargs)
+    dataset = build_datapipeline(urls, transform, batch_size, return_metadata, *args, **kwargs)
     return dataset
 
 
-def build_datapipeline(urls, transform, batch_size, *args, **kwargs):
+def build_datapipeline(urls, transform, batch_size, return_metadata, *args, **kwargs):
     datapipeline = wds.DataPipeline(
         # Infinitely sample shards from the shard list with replacement. Each worker is seeded independently.
         wds.ResampledShards(urls),
         multi_tarfile_samples,  # Extract individual samples from multi-modal tar files
         wds.shuffle(100),  # Shuffle with a buffer of given size
-        wds.decode(zarr_decoder),  # Decode from bytes to PIL images, numpy arrays, etc.
+        (
+            wds.map(zarr_metadata_decoder)
+            if return_metadata
+            else wds.decode(zarr_decoder)  # Decode from bytes to numpy arrays, etc.
+        ),
         wds.map(drop_time_dim),  # Remove time dimension from tensors
         wds.map(remove_extensions),  # Remove "file extensions" from dictionary keys
         (  # Apply transformation
@@ -211,7 +240,7 @@ def extract_modality_names(s):
     """
     Function from https://github.com/apple/ml-4m/blob/main/fourm/data/unified_datasets.py.
     """
-    # Regular expression pattern to match anything enclosed in '{' and '}', and comma separated
+    # Regular expression pattern to match anything enclosed in "{" and "}", and comma separated
     pattern = r"\{([^}]*)\}"
     match = re.search(pattern, s)
     return match.group(1).split(",") if match else []
@@ -259,8 +288,8 @@ def multi_tarfile_samples(
 
     Args:
         src_iter: Iterator over shards that *already brace expanded the shard numbers*,
-            e.g. {'url': 'shard_root_train_[rgb,caption]/00000.tar'}, {'url': 'shard_root_train_[rgb,caption]/00001.tar'}, ...
-            This function will also work when no square braces for multiple modalities are used, e.g. {'url': 'shard_root_train/00000.tar'}, ...
+            e.g. {"url": "shard_root_train_[rgb,caption]/00000.tar"}, {"url": "shard_root_train_[rgb,caption]/00001.tar"}, ...
+            This function will also work when no square braces for multiple modalities are used, e.g. {"url": "shard_root_train/00000.tar"}, ...
             It can be a drop-in replacement for wds.tarfile_samples.
         handler: Function that handles exceptions. If it returns True, the shard is skipped. If it returns False, the function exits.
 
@@ -335,14 +364,14 @@ class Transpose(albumentations.ImageOnlyTransform):
         self.axis = axis
 
     def apply(self, img, **params):
-        return numpy.transpose(img, self.axis)
+        return np.transpose(img, self.axis)
 
     def get_transform_init_args_names(self):
         return "transpose"
 
 
 def default_non_image_transform(array):
-    if hasattr(array, 'dtype') and (array.dtype == float or array.dtype == int):
+    if hasattr(array, "dtype") and (array.dtype == float or array.dtype == int):
         return torch.from_numpy(array)
     else:
         return array
@@ -361,7 +390,6 @@ class MultimodalTransforms:
     def __init__(
             self,
             transforms: dict | albumentations.Compose,
-            shared: bool = True,
             non_image_modalities: list[str] | None = None,
             non_image_transforms: object | None = None,
     ):
@@ -379,14 +407,16 @@ class MultimodalTransforms:
         self.non_image_transforms = non_image_transforms or default_non_image_transform
 
     def __call__(self, data: dict):
-        # albumentations requires a key 'image' and treats all other keys as additional targets
-        image_modality = [k for k in data.keys() if k not in self.non_image_modalities][0]
-        data['image'] = data.pop(image_modality)
+        # albumentations requires a key "image" and treats all other keys as additional targets
+        image_modality = "image" if "image" in data else \
+            [k for k in data.keys() if k not in self.non_image_modalities][0]  # Find an image modality name
+        data["image"] = data.pop(image_modality)  # albumentations expects an input called "image"
         data = self.transforms(**data)
-        data[image_modality] = data.pop('image')
+        data[image_modality] = data.pop("image")
 
-        # Process sequence data which is ignored by albumentations as 'global_label'
+        # Process sequence data which is ignored by albumentations as "global_label"
         for modality in self.non_image_modalities:
-            data[modality] = self.non_image_transforms(data[modality])
+            if modality in data:
+                data[modality] = self.non_image_transforms(data[modality])
 
         return data