Skip to content

The agent engineering platform.

community[patch]: Handle gray scale images in ImageBlobParser (Fixes 30261 and 29586) (#30493)

Philippe PRADOS92189c8b3150
diff --git a/libs/community/langchain_community/document_loaders/parsers/images.py b/libs/community/langchain_community/document_loaders/parsers/images.py
index eb3da4d2a1..3d977aae97 100644
--- a/libs/community/langchain_community/document_loaders/parsers/images.py
+++ b/libs/community/langchain_community/document_loaders/parsers/images.py
@@ -53,7 +53,11 @@ class BaseImageBlobParser(BaseBlobParser):
with blob.as_bytes_io() as buf:
if blob.mimetype == "application/x-npy":
- img = Img.fromarray(numpy.load(buf))
+ array = numpy.load(buf)
+ if array.ndim == 3 and array.shape[2] == 1: # Grayscale image
+ img = Img.fromarray(numpy.squeeze(array, axis=2), mode="L")
+ else:
+ img = Img.fromarray(array)
else:
img = Img.open(buf)
content = self._analyze_image(img)
diff --git a/libs/community/tests/integration_tests/document_loaders/parsers/test_images.py b/libs/community/tests/integration_tests/document_loaders/parsers/test_images.py
index e6d71fae69..e5c6e372b8 100644
--- a/libs/community/tests/integration_tests/document_loaders/parsers/test_images.py
+++ b/libs/community/tests/integration_tests/document_loaders/parsers/test_images.py
@@ -1,7 +1,9 @@
import re
+from io import BytesIO
from pathlib import Path
from typing import Any, Type
+import numpy as np
import pytest
from langchain_core.documents.base import Blob
from langchain_core.language_models import FakeMessagesListChatModel
@@ -18,12 +20,13 @@ building_image = Blob.from_path(path_base / "examples/building.jpg")
text_image = Blob.from_path(path_base / "examples/text.png")
page_image = Blob.from_path(path_base / "examples/page.png")
+_re_in_image = r"(?ms).*MAKE.*TEXT.*STAND.*OUT.*FROM.*"
+
@pytest.mark.parametrize(
"blob,body",
[
- (building_image, ""),
- (text_image, r"(?ms).*MAKE.*TEXT.*STAND.*OUT.*FROM.*BACKGROUNDS.*"),
+ (Blob.from_path(path_base / "examples/text-gray.png"), _re_in_image),
],
)
@pytest.mark.parametrize(
@@ -58,3 +61,39 @@ def test_image_parser_with_differents_files(
documents = list(blob_loader(**kw).lazy_parse(blob))
assert len(documents) == 1
assert re.compile(body).match(documents[0].page_content)
+
+
+ "blob_loader,kw",
+ [
+ (RapidOCRBlobParser, {}),
+ (TesseractBlobParser, {}),
+ (
+ LLMImageBlobParser,
+ {
+ "model": FakeMessagesListChatModel(
+ responses=[
+ ChatMessage(
+ id="ai1",
+ role="system",
+ content="A building. MAKE TEXT STAND OUT FROM BACKGROUNDS",
+ ),
+ ]
+ )
+ },
+ ),
+ ],
+)
+def test_image_parser_with_numpy(
+ blob_loader: Type,
+ kw: dict[str, Any],
+) -> None:
+ gray_image = np.empty(shape=(412, 1652, 1))
+ with BytesIO() as buffer:
+ np.save(buffer, gray_image)
+ buffer.seek(0)
+ npy_bytes = buffer.getvalue()
+
+ blob = Blob.from_data(npy_bytes, mime_type="application/x-npy")
+ documents = list(blob_loader(**kw).lazy_parse(blob))
+ assert len(documents) == 1