Added module which does OCR for PDF, pdf-ocr in a separate package (#324)

(has a lot of dependencies).  Uses Tesseract.

Containerfile

@@ -45,6 +45,7 @@ COPY trustgraph-vertexai/ /root/build/trustgraph-vertexai/
 COPY trustgraph-bedrock/ /root/build/trustgraph-bedrock/
 COPY trustgraph-embeddings-hf/ /root/build/trustgraph-embeddings-hf/
 COPY trustgraph-cli/ /root/build/trustgraph-cli/
+COPY trustgraph-ocr/ /root/build/trustgraph-ocr/
 
 WORKDIR /root/build/
 
@@ -54,6 +55,7 @@ RUN pip3 wheel -w /root/wheels/ --no-deps ./trustgraph-vertexai/
 RUN pip3 wheel -w /root/wheels/ --no-deps ./trustgraph-bedrock/
 RUN pip3 wheel -w /root/wheels/ --no-deps ./trustgraph-embeddings-hf/
 RUN pip3 wheel -w /root/wheels/ --no-deps ./trustgraph-cli/
+RUN pip3 wheel -w /root/wheels/ --no-deps ./trustgraph-ocr/
 
 RUN ls /root/wheels
 
@@ -72,6 +74,7 @@ RUN \
     pip3 install /root/wheels/trustgraph_bedrock-* && \
     pip3 install /root/wheels/trustgraph_embeddings_hf-* && \
     pip3 install /root/wheels/trustgraph_cli-* && \
+    pip3 install /root/wheels/trustgraph_ocr-* && \
     pip3 cache purge && \
     rm -rf /root/wheels
 

Makefile

@@ -16,6 +16,7 @@ wheels:
 	pip3 wheel --no-deps --wheel-dir dist trustgraph-bedrock/
 	pip3 wheel --no-deps --wheel-dir dist trustgraph-embeddings-hf/
 	pip3 wheel --no-deps --wheel-dir dist trustgraph-cli/
+	pip3 wheel --no-deps --wheel-dir dist trustgraph-ocr/
 
 packages: update-package-versions
 	rm -rf dist/
@@ -26,6 +27,7 @@ packages: update-package-versions
 	cd trustgraph-bedrock && python3 setup.py sdist --dist-dir ../dist/
 	cd trustgraph-embeddings-hf && python3 setup.py sdist --dist-dir ../dist/
 	cd trustgraph-cli && python3 setup.py sdist --dist-dir ../dist/
+	cd trustgraph-ocr && python3 setup.py sdist --dist-dir ../dist/
 
 pypi-upload:
 	twine upload dist/*-${VERSION}.*
@@ -41,6 +43,7 @@ update-package-versions:
 	echo __version__ = \"${VERSION}\" > trustgraph-bedrock/trustgraph/bedrock_version.py
 	echo __version__ = \"${VERSION}\" > trustgraph-embeddings-hf/trustgraph/embeddings_hf_version.py
 	echo __version__ = \"${VERSION}\" > trustgraph-cli/trustgraph/cli_version.py
+	echo __version__ = \"${VERSION}\" > trustgraph-ocr/trustgraph/ocr_version.py
 	echo __version__ = \"${VERSION}\" > trustgraph/trustgraph/trustgraph_version.py
 
 container: update-package-versions
@@ -54,6 +57,12 @@ container: update-package-versions
 	    -t ${CONTAINER_BASE}/trustgraph-vertexai:${VERSION} .
 	${DOCKER} build -f containers/Containerfile.hf \
 	    -t ${CONTAINER_BASE}/trustgraph-hf:${VERSION} .
+	${DOCKER} build -f containers/Containerfile.ocr \
+	    -t ${CONTAINER_BASE}/trustgraph-ocr:${VERSION} .
+
+container.ocr:
+	${DOCKER} build -f containers/Containerfile.ocr \
+	    -t ${CONTAINER_BASE}/trustgraph-ocr:${VERSION} .
 
 push:
 	${DOCKER} push ${CONTAINER_BASE}/trustgraph-base:${VERSION}
@@ -61,6 +70,7 @@ push:
 	${DOCKER} push ${CONTAINER_BASE}/trustgraph-bedrock:${VERSION}
 	${DOCKER} push ${CONTAINER_BASE}/trustgraph-vertexai:${VERSION}
 	${DOCKER} push ${CONTAINER_BASE}/trustgraph-hf:${VERSION}
+	${DOCKER} push ${CONTAINER_BASE}/trustgraph-ocr:${VERSION}
 
 clean:
 	rm -rf wheels/

containers/Containerfile.ocr

@@ -0,0 +1,48 @@
+
+# ----------------------------------------------------------------------------
+# Build an AI container.  This does the torch install which is huge, and I
+# like to avoid re-doing this.
+# ----------------------------------------------------------------------------
+
+FROM docker.io/fedora:40 AS base
+
+ENV PIP_BREAK_SYSTEM_PACKAGES=1
+
+RUN dnf install -y python3 python3-pip python3-wheel python3-aiohttp \
+    python3-rdflib tesseract poppler poppler-utils
+
+RUN pip3 install --no-cache-dir pytesseract pulsar-client==3.5.0
+
+# ----------------------------------------------------------------------------
+# Build a container which contains the built Python packages.  The build
+# creates a bunch of left-over cruft, a separate phase means this is only
+# needed to support package build
+# ----------------------------------------------------------------------------
+
+FROM base AS build
+
+COPY trustgraph-base/ /root/build/trustgraph-base/
+COPY trustgraph-tesseract/ /root/build/trustgraph-tesseract/
+
+WORKDIR /root/build/
+
+RUN pip3 wheel -w /root/wheels/ --no-deps ./trustgraph-base/
+RUN pip3 wheel -w /root/wheels/ --no-deps ./trustgraph-tesseract/
+
+RUN ls /root/wheels
+
+# ----------------------------------------------------------------------------
+# Finally, the target container.  Start with base and add the package.
+# ----------------------------------------------------------------------------
+
+FROM base
+
+COPY --from=build /root/wheels /root/wheels
+
+RUN \
+    pip3 install --no-cache-dir /root/wheels/trustgraph_base-* && \
+    pip3 install --no-cache-dir /root/wheels/trustgraph_ocr-* && \
+    rm -rf /root/wheels
+
+WORKDIR /
+

trustgraph-ocr/README.md

@@ -0,0 +1 @@
+See https://trustgraph.ai/

trustgraph-ocr/scripts/pdf-ocr

@@ -0,0 +1,6 @@
+#!/usr/bin/env python3
+
+from trustgraph.decoding.ocr import run
+
+run()
+

trustgraph-ocr/setup.py

@@ -0,0 +1,47 @@
+import setuptools
+import os
+import importlib
+
+with open("README.md", "r") as fh:
+    long_description = fh.read()
+
+# Load a version number module
+spec = importlib.util.spec_from_file_location(
+    'version', 'trustgraph/ocr_version.py'
+)
+version_module = importlib.util.module_from_spec(spec)
+spec.loader.exec_module(version_module)
+
+version = version_module.__version__
+
+setuptools.setup(
+    name="trustgraph-ocr",
+    version=version,
+    author="trustgraph.ai",
+    author_email="security@trustgraph.ai",
+    description="TrustGraph provides a means to run a pipeline of flexible AI processing components in a flexible means to achieve a processing pipeline.",
+    long_description=long_description,
+    long_description_content_type="text/markdown",
+    url="https://github.com/trustgraph-ai/trustgraph",
+    packages=setuptools.find_namespace_packages(
+        where='./',
+    ),
+    classifiers=[ 
+        "Programming Language :: Python :: 3",
+        "License :: OSI Approved :: GNU General Public License v3 or later (GPLv3+)",
+        "Operating System :: OS Independent",
+    ],
+    python_requires='>=3.8',
+    download_url = "https://github.com/trustgraph-ai/trustgraph/archive/refs/tags/v" + version + ".tar.gz",
+    install_requires=[
+        "trustgraph-base>=0.21,<0.22",
+        "pulsar-client",
+        "prometheus-client",
+        "boto3",
+        "pdf2image",
+        "pytesseract",
+    ],
+    scripts=[
+        "scripts/pdf-ocr",
+    ]
+)

trustgraph-ocr/trustgraph/decoding/ocr/__init__.py

@@ -0,0 +1,3 @@
+
+from . pdf_decoder import *
+

trustgraph-ocr/trustgraph/decoding/ocr/__main__.py

@@ -0,0 +1,7 @@
+#!/usr/bin/env python3
+
+from . pdf_decoder import run
+
+if __name__ == '__main__':
+    run()
+

trustgraph-ocr/trustgraph/decoding/ocr/pdf_decoder.py

@@ -0,0 +1,83 @@
+
+"""
+Simple decoder, accepts PDF documents on input, outputs pages from the
+PDF document as text as separate output objects.
+"""
+
+import tempfile
+import base64
+import pytesseract
+from pdf2image import convert_from_bytes
+
+from ... schema import Document, TextDocument, Metadata
+from ... schema import document_ingest_queue, text_ingest_queue
+from ... log_level import LogLevel
+from ... base import ConsumerProducer
+
+module = ".".join(__name__.split(".")[1:-1])
+
+default_input_queue = document_ingest_queue
+default_output_queue = text_ingest_queue
+default_subscriber = module
+
+class Processor(ConsumerProducer):
+
+    def __init__(self, **params):
+
+        input_queue = params.get("input_queue", default_input_queue)
+        output_queue = params.get("output_queue", default_output_queue)
+        subscriber = params.get("subscriber", default_subscriber)
+
+        super(Processor, self).__init__(
+            **params | {
+                "input_queue": input_queue,
+                "output_queue": output_queue,
+                "subscriber": subscriber,
+                "input_schema": Document,
+                "output_schema": TextDocument,
+            }
+        )
+
+        print("PDF OCR inited")
+
+    async def handle(self, msg):
+
+        print("PDF message received")
+
+        v = msg.value()
+
+        print(f"Decoding {v.metadata.id}...", flush=True)
+
+        blob = base64.b64decode(v.data)
+
+        pages = convert_from_bytes(blob)
+
+        for ix, page in enumerate(pages):
+
+            try:
+                text = pytesseract.image_to_string(page, lang='eng')
+            except Exception as e:
+                print(f"Page did not OCR: {e}")
+                continue
+
+            r = TextDocument(
+                metadata=v.metadata,
+                text=text.encode("utf-8"),
+            )
+
+            await self.send(r)
+
+        print("Done.", flush=True)
+
+    @staticmethod
+    def add_args(parser):
+
+        ConsumerProducer.add_args(
+            parser, default_input_queue, default_subscriber,
+            default_output_queue,
+        )
+
+def run():
+
+    Processor.launch(module, __doc__)
+