stuff

python/pyproject.toml

@@ -57,6 +57,7 @@ tests = [
     "duckdb",
     "pytz",
     "polars>=0.19",
+    "pillow",
 ]
 dev = ["ruff", "pre-commit"]
 docs = [

python/python/lancedb/embeddings/open_clip.py

@@ -126,6 +126,10 @@ class OpenClipEmbeddings(EmbeddingFunction):
         """
         Issue concurrent requests to retrieve the image data
         """
+        return [
+            self.generate_image_embedding(image) for image in tqdm(images)
+        ]
+
         with concurrent.futures.ThreadPoolExecutor() as executor:
             futures = [
                 executor.submit(self.generate_image_embedding, image)

python/python/lancedb/pydantic.py

@@ -16,6 +16,7 @@
 from __future__ import annotations
 
 import inspect
+import io
 import sys
 import types
 from abc import ABC, abstractmethod
@@ -26,7 +27,9 @@ from typing import (
     Callable,
     Dict,
     Generator,
+    Iterable,
     List,
+    Tuple,
     Type,
     Union,
     _GenericAlias,
@@ -36,19 +39,30 @@ import numpy as np
 import pyarrow as pa
 import pydantic
 import semver
+from lance.arrow import (
+    EncodedImageType,
+)
+from lance.util import _check_huggingface
+from pydantic.fields import FieldInfo
+from pydantic_core import core_schema
+
+from .util import attempt_import_or_raise
 
 PYDANTIC_VERSION = semver.Version.parse(pydantic.__version__)
-try:
-    from pydantic_core import CoreSchema, core_schema
-except ImportError:
-    if PYDANTIC_VERSION >= (2,):
-        raise
 
 if TYPE_CHECKING:
     from pydantic.fields import FieldInfo
 
     from .embeddings import EmbeddingFunctionConfig
 
+    try:
+        from pydantic import GetJsonSchemaHandler
+        from pydantic.json_schema import JsonSchemaValue
+        from pydantic_core import CoreSchema
+    except ImportError:
+        if PYDANTIC_VERSION >= (2,):
+            raise
+
 
 class FixedSizeListMixin(ABC):
     @staticmethod
@@ -123,7 +137,7 @@ def Vector(
         @classmethod
         def __get_pydantic_core_schema__(
             cls, _source_type: Any, _handler: pydantic.GetCoreSchemaHandler
-        ) -> CoreSchema:
+        ) -> "CoreSchema":
             return core_schema.no_info_after_validator_function(
                 cls,
                 core_schema.list_schema(
@@ -181,25 +195,118 @@ def _py_type_to_arrow_type(py_type: Type[Any], field: FieldInfo) -> pa.DataType:
     elif getattr(py_type, "__origin__", None) in (list, tuple):
         child = py_type.__args__[0]
         return pa.list_(_py_type_to_arrow_type(child, field))
+    elif _safe_is_huggingface_image():
+        import datasets
+        
     raise TypeError(
         f"Converting Pydantic type to Arrow Type: unsupported type {py_type}."
     )
 
 
+class ImageMixin(ABC):
+    @staticmethod
+    @abstractmethod
+    def value_arrow_type() -> pa.DataType:
+        raise NotImplementedError
+
+
+def EncodedImage():
+    attempt_import_or_raise("PIL", "pillow or pip install lancedb[embeddings]")
+    import PIL.Image
+
+    class EncodedImage(bytes, ImageMixin):
+        """Pydantic type for inlined images.
+
+        !!! warning
+            Experimental feature.
+
+        Examples
+        --------
+
+        >>> import pydantic
+        >>> from lancedb.pydantic import EncodedImage
+        ...
+        >>> class MyModel(pydantic.BaseModel):
+        ...     image: EncodedImage()
+        >>> schema = pydantic_to_schema(MyModel)
+        >>> assert schema == pa.schema([
+        ...     pa.field("image", pa.binary(), False)
+        ... ])
+        """
+
+        def __repr__(self):
+            return "EncodedImage()"
+
+        @staticmethod
+        def value_arrow_type() -> pa.DataType:
+            return EncodedImageType()
+
+        @classmethod
+        def __get_pydantic_core_schema__(
+            cls, _source_type: Any, _handler: pydantic.GetCoreSchemaHandler
+        ) -> "CoreSchema":
+            from_bytes_schema = core_schema.bytes_schema()
+
+            return core_schema.json_or_python_schema(
+                json_schema=from_bytes_schema,
+                python_schema=core_schema.union_schema(
+                    [
+                        core_schema.is_instance_schema(PIL.Image.Image),
+                        from_bytes_schema,
+                    ]
+                ),
+                serialization=core_schema.plain_serializer_function_ser_schema(
+                    lambda instance: cls.validate(instance)
+                ),
+            )
+
+        @classmethod
+        def __get_pydantic_json_schema__(
+            cls, _core_schema: "CoreSchema", handler: "GetJsonSchemaHandler"
+        ) -> "JsonSchemaValue":
+            return handler(core_schema.bytes_schema())
+
+        @classmethod
+        def __get_validators__(cls) -> Generator[Callable, None, None]:
+            yield cls.validate
+
+        # For pydantic v2
+        @classmethod
+        def validate(cls, v):
+            if isinstance(v, bytes):
+                return v
+            if isinstance(v, PIL.Image.Image):
+                with io.BytesIO() as output:
+                    v.save(output, format=v.format)
+                    return output.getvalue()
+            raise TypeError(
+                "EncodedImage can take bytes or PIL.Image.Image "
+                f"as input but got {type(v)}"
+            )
+
+        if PYDANTIC_VERSION < (2, 0):
+
+            @classmethod
+            def __modify_schema__(cls, field_schema: Dict[str, Any]):
+                field_schema["type"] = "string"
+                field_schema["format"] = "binary"
+
+    return EncodedImage
+
+
 if PYDANTIC_VERSION.major < 2:
-
+    def _safe_get_fields(model: pydantic.BaseModel):
-    def _pydantic_model_to_fields(model: pydantic.BaseModel) -> List[pa.Field]:
+        return model.__fields__
-        return [
-            _pydantic_to_field(name, field) for name, field in model.__fields__.items()
-        ]
-
 else:
+    def _safe_get_fields(model: pydantic.BaseModel):    
+        return model.model_fields
 
-    def _pydantic_model_to_fields(model: pydantic.BaseModel) -> List[pa.Field]:
+
-        return [
+def _pydantic_model_to_fields(model: pydantic.BaseModel) -> List[pa.Field]:
-            _pydantic_to_field(name, field)
+    return [
-            for name, field in model.model_fields.items()
+        _pydantic_to_field(name, field)
-        ]
+        for name, field in _safe_get_fields(model).items()
+    ]
 
 
 def _pydantic_to_arrow_type(field: FieldInfo) -> pa.DataType:
@@ -230,6 +337,9 @@ def _pydantic_to_arrow_type(field: FieldInfo) -> pa.DataType:
             return pa.struct(fields)
         elif issubclass(field.annotation, FixedSizeListMixin):
             return pa.list_(field.annotation.value_arrow_type(), field.annotation.dim())
+        elif issubclass(field.annotation, ImageMixin):
+            return field.annotation.value_arrow_type()
+
     return _py_type_to_arrow_type(field.annotation, field)
 
 
@@ -335,13 +445,7 @@ class LanceModel(pydantic.BaseModel):
         """
         Get the field names of this model.
         """
-        return list(cls.safe_get_fields().keys())
+        return list(_safe_get_fields(cls).keys())
-
-    @classmethod
-    def safe_get_fields(cls):
-        if PYDANTIC_VERSION.major < 2:
-            return cls.__fields__
-        return cls.model_fields
 
     @classmethod
     def parse_embedding_functions(cls) -> List["EmbeddingFunctionConfig"]:
@@ -351,14 +455,16 @@ class LanceModel(pydantic.BaseModel):
         from .embeddings import EmbeddingFunctionConfig
 
         vec_and_function = []
-        for name, field_info in cls.safe_get_fields().items():
+        def get_vector_column(name, field_info):
-            func = get_extras(field_info, "vector_column_for")
+            fun = get_extras(field_info, "vector_column_for")
             if func is not None:
                 vec_and_function.append([name, func])
+        visit_fields(_safe_get_fields(cls).items(), get_vector_column)
 
         configs = []
+        # find the source columns for each one
         for vec, func in vec_and_function:
-            for source, field_info in cls.safe_get_fields().items():
+            def get_source_column(source, field_info):
                 src_func = get_extras(field_info, "source_column_for")
                 if src_func is func:
                     # note we can't use == here since the function is a pydantic
@@ -371,20 +477,48 @@ class LanceModel(pydantic.BaseModel):
                             source_column=source, vector_column=vec, function=func
                         )
                     )
+            visit_fields(_safe_get_fields(cls).items(), get_source_column)
         return configs
+    
+
+def visit_fields(fields: Iterable[Tuple[str, FieldInfo]], 
+                 visitor: Callable[[str, FieldInfo], Any]):
+    """
+    Visit all the leaf fields in a Pydantic model.
+
+    Parameters
+    ----------
+    fields : Iterable[Tuple(str, FieldInfo)]
+        The fields to visit.
+    visitor : Callable[[str, FieldInfo], Any]
+        The visitor function.
+    """
+    for name, field_info in fields:
+        # if the field is a pydantic model then
+        # visit all subfields
+        if (isinstance(getattr(field_info, "annotation"), type) 
+            and issubclass(field_info.annotation, pydantic.BaseModel)):
+            visit_fields(_safe_get_fields(field_info.annotation).items(), 
+                           _add_prefix(visitor, name))
+        else:
+            visitor(name, field_info)
 
 
-def get_extras(field_info: FieldInfo, key: str) -> Any:
+def _add_prefix(visitor: Callable[[str, FieldInfo], Any], prefix: str) -> Callable[[str, FieldInfo], Any]:
-    """
+    def prefixed_visitor(name: str, field: FieldInfo):
-    Get the extra metadata from a Pydantic FieldInfo.
+        return visitor(f"{prefix}.{name}", field)
-    """
+    return prefixed_visitor
-    if PYDANTIC_VERSION.major >= 2:
-        return (field_info.json_schema_extra or {}).get(key)
-    return (field_info.field_info.extra or {}).get("json_schema_extra", {}).get(key)
 
 
 if PYDANTIC_VERSION.major < 2:
 
+    def get_extras(field_info: FieldInfo, key: str) -> Any:
+        """
+        Get the extra metadata from a Pydantic FieldInfo.
+        """
+        return (field_info.field_info.extra or {}).get("json_schema_extra", {}).get(key)
+
+
     def model_to_dict(model: pydantic.BaseModel) -> Dict[str, Any]:
         """
         Convert a Pydantic model to a dictionary.
@@ -393,6 +527,13 @@ if PYDANTIC_VERSION.major < 2:
 
 else:
 
+    def get_extras(field_info: FieldInfo, key: str) -> Any:
+        """
+        Get the extra metadata from a Pydantic FieldInfo.
+        """
+        return (field_info.json_schema_extra or {}).get(key)
+    
+
     def model_to_dict(model: pydantic.BaseModel) -> Dict[str, Any]:
         """
         Convert a Pydantic model to a dictionary.

python/python/lancedb/table.py

@@ -37,6 +37,7 @@ import pyarrow as pa
 import pyarrow.compute as pc
 import pyarrow.fs as pa_fs
 from lance import LanceDataset
+from lance.dependencies import _check_for_hugging_face
 from lance.vector import vec_to_table
 
 from .common import DATA, VEC, VECTOR_COLUMN_NAME
@@ -74,7 +75,16 @@ def _sanitize_data(
     on_bad_vectors: str,
     fill_value: Any,
 ):
-    if isinstance(data, list):
+    import pdb; pdb.set_trace()
+    if _check_for_hugging_face(data):
+        # Huggingface datasets
+        import datasets
+
+        if isinstance(data, datasets.Dataset):
+            if schema is None:
+                schema = data.features.arrow_schema
+            data = data.data.to_batches()
+    elif isinstance(data, list):
         # convert to list of dict if data is a bunch of LanceModels
         if isinstance(data[0], LanceModel):
             schema = data[0].__class__.to_arrow_schema()
@@ -136,11 +146,10 @@ def _to_record_batch_generator(
     data: Iterable, schema, metadata, on_bad_vectors, fill_value
 ):
     for batch in data:
-        if not isinstance(batch, pa.RecordBatch):
+        if isinstance(batch, pa.RecordBatch):
-            table = _sanitize_data(batch, schema, metadata, on_bad_vectors, fill_value)
+            batch = pa.Table.from_batches([batch])
-            for batch in table.to_batches():
+        table = _sanitize_data(batch, schema, metadata, on_bad_vectors, fill_value)
-                yield batch
+        for batch in table.to_batches():
-        else:
             yield batch
 
 

python/python/tests/test_pydantic.py

@@ -12,17 +12,27 @@
 #  limitations under the License.
 
 
+import io
 import json
+import os
 import sys
 from datetime import date, datetime
+from pathlib import Path
 from typing import List, Optional, Tuple
 
 import pyarrow as pa
 import pydantic
 import pytest
-from lancedb.pydantic import PYDANTIC_VERSION, LanceModel, Vector, pydantic_to_schema
 from pydantic import Field
 
+from lancedb.pydantic import (
+    PYDANTIC_VERSION,
+    EncodedImage,
+    LanceModel,
+    Vector,
+    pydantic_to_schema,
+)
+
 
 @pytest.mark.skipif(
     sys.version_info < (3, 9),
@@ -243,3 +253,23 @@ def test_lance_model():
 
     t = TestModel()
     assert t == TestModel(vec=[0.0] * 16, li=[1, 2, 3])
+
+
+def test_schema_with_images():
+    pytest.importorskip("PIL")
+    import PIL.Image
+
+    class TestModel(LanceModel):
+        img: EncodedImage()
+
+    img_path = Path(os.path.dirname(__file__)) / "images/1.png"
+    with open(img_path, "rb") as f:
+        img_bytes = f.read()
+
+    m1 = TestModel(img=PIL.Image.open(img_path))
+    m2 = TestModel(img=img_bytes)
+
+    def tobytes(m):
+        return PIL.Image.open(io.BytesIO(m.model_dump()["img"])).tobytes()
+
+    assert tobytes(m1) == tobytes(m2)

python/python/tests/test_table.py

@@ -12,6 +12,8 @@
 #  limitations under the License.
 
 import functools
+import io
+import os
 from copy import copy
 from datetime import date, datetime, timedelta
 from pathlib import Path
@@ -20,19 +22,21 @@ from typing import List
 from unittest.mock import PropertyMock, patch
 
 import lance
-import lancedb
 import numpy as np
 import pandas as pd
 import polars as pl
 import pyarrow as pa
 import pytest
 import pytest_asyncio
+from lance.arrow import EncodedImageType
+from pydantic import BaseModel
+
+import lancedb
 from lancedb.conftest import MockTextEmbeddingFunction
 from lancedb.db import AsyncConnection, LanceDBConnection
 from lancedb.embeddings import EmbeddingFunctionConfig, EmbeddingFunctionRegistry
-from lancedb.pydantic import LanceModel, Vector
+from lancedb.pydantic import EncodedImage, LanceModel, Vector
 from lancedb.table import LanceTable
-from pydantic import BaseModel
 
 
 class MockDB:
@@ -108,6 +112,7 @@ def test_create_table(db):
             pa.field("vector", pa.list_(pa.float32(), 2)),
             pa.field("item", pa.string()),
             pa.field("price", pa.float32()),
+            pa.field("encoded_image", EncodedImageType()),
         ]
     )
     expected = pa.Table.from_arrays(
@@ -115,13 +120,26 @@ def test_create_table(db):
             pa.FixedSizeListArray.from_arrays(pa.array([3.1, 4.1, 5.9, 26.5]), 2),
             pa.array(["foo", "bar"]),
             pa.array([10.0, 20.0]),
+            pa.ExtensionArray.from_storage(
+                EncodedImageType(), pa.array([b"foo", b"bar"], pa.binary())
+            ),
         ],
         schema=schema,
     )
     data = [
         [
-            {"vector": [3.1, 4.1], "item": "foo", "price": 10.0},
+            {
-            {"vector": [5.9, 26.5], "item": "bar", "price": 20.0},
+                "vector": [3.1, 4.1],
+                "item": "foo",
+                "price": 10.0,
+                "encoded_image": b"foo",
+            },
+            {
+                "vector": [5.9, 26.5],
+                "item": "bar",
+                "price": 20.0,
+                "encoded_image": b"bar",
+            },
         ]
     ]
     df = pd.DataFrame(data[0])
@@ -1025,3 +1043,22 @@ async def test_time_travel(db_async: AsyncConnection):
     # Can't use restore if not checked out
     with pytest.raises(ValueError, match="checkout before running restore"):
         await table.restore()
+
+        
+def test_add_image(tmp_path):
+    pytest.importorskip("PIL")
+    import PIL.Image
+
+    db = lancedb.connect(tmp_path)
+
+    class TestModel(LanceModel):
+        img: EncodedImage()
+
+    img_path = Path(os.path.dirname(__file__)) / "images/1.png"
+    m1 = TestModel(img=PIL.Image.open(img_path))
+
+    def tobytes(m):
+        return PIL.Image.open(io.BytesIO(m.model_dump()["img"])).tobytes()
+
+    table = LanceTable.create(db, "my_table", schema=TestModel)
+    table.add([m1])