Reset OCR engine to tesseract (#78)

##

Author: alimaredia

kubeflow-pipelines/common/components.py

@@ -175,7 +175,7 @@ def download_docling_models(
     Download Docling models based on pipeline type and configuration.
 
     This unified component handles model downloading for different pipeline types:
-    - standard : Download traditional Docling models (layout, tableformer, easyocr)
+    - standard : Download traditional Docling models (layout, tableformer)
     - vlm : Download Docling VLM models (smolvlm, smoldocling) for local inference
              When remote_model_endpoint_enabled=True, downloads minimal models for remote inference
 
@@ -200,7 +200,7 @@ def download_docling_models(
             progress=True,
             with_layout=True,
             with_tableformer=True,
-            with_easyocr=True,
+            with_easyocr=False,
         )
     elif pipeline_type == "vlm" and remote_model_endpoint_enabled:
         # VLM pipeline with remote model endpoint: Download minimal required models

kubeflow-pipelines/common/constants.py

@@ -2,8 +2,8 @@
 
 # Base container images used across all Docling Kubeflow Pipelines
 PYTHON_BASE_IMAGE = os.getenv(
-    "PYTHON_BASE_IMAGE", "quay.io/amaredia/aipcc-docling-image"
+    "PYTHON_BASE_IMAGE", "registry.access.redhat.com/ubi9/python-311:9.6-1755074620"
 )
 DOCLING_BASE_IMAGE = os.getenv(
-    "DOCLING_BASE_IMAGE", "quay.io/amaredia/aipcc-docling-image"
+    "DOCLING_BASE_IMAGE", "quay.io/fabianofranz/docling-ubi9:2.54.0"
 )

kubeflow-pipelines/docling-standard/README.md

@@ -18,7 +18,7 @@ The following configuration options are available as KFP parameters when you _Cr
 - `docling_image_export_mode`: Image export mode for the document. In `embedded` mode, the image is embedded as base64 encoded string. With `placeholder`, only the position of the image is marked in the output. In `referenced` mode, the image is exported in PNG format and referenced from the main exported document.
 - `docling_num_threads`: Number of threads to be used internally by the Docling engine.
 - `docling_ocr`: If enabled, the bitmap content will be processed using OCR.
-- `docling_ocr_engine`: The OCR engine to use. Current values are: `easyocr`.
+- `docling_ocr_engine`: The OCR engine to use. Current values are: `tesseract_cli`.
 - `docling_pdf_backend`: The PDF backend to use. `pypdfium2`, `dlparse_v1`, `dlparse_v2`, or `dlparse_v4`.
 - `docling_table_mode`: The mode to use in the table structure model. `accurate` or `fast`.
 - `docling_timeout_per_document`: Timeout for each single document conversion.

kubeflow-pipelines/docling-standard/standard_components.py

@@ -24,7 +24,7 @@ def docling_convert_standard(
     timeout_per_document: int = 300,
     ocr: bool = True,
     force_ocr: bool = False,
-    ocr_engine: str = "easyocr",
+    ocr_engine: str = "tesseract_cli",
     allow_external_plugins: bool = False,
     enrich_code: bool = False,
     enrich_formula: bool = False,

kubeflow-pipelines/docling-standard/standard_convert_pipeline.py

@@ -33,7 +33,7 @@ def convert_pipeline(
     docling_timeout_per_document: int = 300,
     docling_ocr: bool = True,
     docling_force_ocr: bool = False,
-    docling_ocr_engine: str = "easyocr",
+    docling_ocr_engine: str = "tesseract_cli",
     docling_allow_external_plugins: bool = False,
     docling_enrich_code: bool = False,
     docling_enrich_formula: bool = False,

kubeflow-pipelines/docling-standard/standard_convert_pipeline_compiled.yaml

@@ -14,7 +14,7 @@
 #    docling_image_export_mode: str [Default: 'embedded']
 #    docling_num_threads: int [Default: 4.0]
 #    docling_ocr: bool [Default: True]
-#    docling_ocr_engine: str [Default: 'easyocr']
+#    docling_ocr_engine: str [Default: 'tesseract_cli']
 #    docling_pdf_backend: str [Default: 'dlparse_v4']
 #    docling_table_mode: str [Default: 'accurate']
 #    docling_timeout_per_document: int [Default: 300.0]
@@ -158,7 +158,7 @@ components:
           isOptional: true
           parameterType: BOOLEAN
         ocr_engine:
-          defaultValue: easyocr
+          defaultValue: tesseract_cli
           description: Engine to use for OCR.
           isOptional: true
           parameterType: STRING
@@ -383,7 +383,7 @@ deploymentSpec:
           \ for path in input_path_p.glob(\"*.pdf\")]\n    all_splits = [all_pdfs[i::num_splits]\
           \ for i in range(num_splits)]\n    filled_splits = list(filter(None, all_splits))\n\
           \    return filled_splits\n\n"
-        image: quay.io/amaredia/aipcc-docling-image
+        image: registry.access.redhat.com/ubi9/python-311:9.6-1755074620
     exec-docling-chunk:
       container:
         args:
@@ -497,7 +497,7 @@ deploymentSpec:
           \            f\"docling-chunk: skipped {len(skipped_files)} invalid files:\"\
           ,\n            flush=True,\n        )\n        for filename, reason in skipped_files:\n\
           \            print(f\"  - {filename}: {reason}\", flush=True)\n\n"
-        image: quay.io/amaredia/aipcc-docling-image
+        image: quay.io/fabianofranz/docling-ubi9:2.54.0
         resources:
           cpuLimit: 2.0
           cpuRequest: 0.25
@@ -539,7 +539,7 @@ deploymentSpec:
           \    image_export_mode: str = \"embedded\",\n    table_mode: str = \"accurate\"\
           ,\n    num_threads: int = 4,\n    timeout_per_document: int = 300,\n   \
           \ ocr: bool = True,\n    force_ocr: bool = False,\n    ocr_engine: str =\
-          \ \"easyocr\",\n    allow_external_plugins: bool = False,\n    enrich_code:\
+          \ \"tesseract_cli\",\n    allow_external_plugins: bool = False,\n    enrich_code:\
           \ bool = False,\n    enrich_formula: bool = False,\n    enrich_picture_classes:\
           \ bool = False,\n    enrich_picture_description: bool = False,\n):\n   \
           \ \"\"\"\n    Convert a list of PDF files to JSON and Markdown using Docling\
@@ -634,7 +634,7 @@ deploymentSpec:
           \ flush=True)\n        result.document.save_as_markdown(\n            output_md_path,\
           \ image_mode=ImageRefMode(image_export_mode)\n        )\n\n    print(\"\
           docling-standard-convert: done\", flush=True)\n\n"
-        image: quay.io/amaredia/aipcc-docling-image
+        image: quay.io/fabianofranz/docling-ubi9:2.54.0
         resources:
           cpuLimit: 4.0
           cpuRequest: 0.5
@@ -675,8 +675,8 @@ deploymentSpec:
           \ bool = False,\n):\n    \"\"\"\n    Download Docling models based on pipeline\
           \ type and configuration.\n\n    This unified component handles model downloading\
           \ for different pipeline types:\n    - standard : Download traditional Docling\
-          \ models (layout, tableformer, easyocr)\n    - vlm : Download Docling VLM\
-          \ models (smolvlm, smoldocling) for local inference\n             When remote_model_endpoint_enabled=True,\
+          \ models (layout, tableformer)\n    - vlm : Download Docling VLM models\
+          \ (smolvlm, smoldocling) for local inference\n             When remote_model_endpoint_enabled=True,\
           \ downloads minimal models for remote inference\n\n    Args:\n        output_path:\
           \ Path to the output directory for Docling models\n        pipeline_type:\
           \ Type of pipeline (standard, vlm)\n        remote_model_endpoint_enabled:\
@@ -688,9 +688,9 @@ deploymentSpec:
           \n    if pipeline_type == \"standard\":\n        # Standard pipeline: download\
           \ traditional models\n        download_models(\n            output_dir=output_path_p,\n\
           \            progress=True,\n            with_layout=True,\n           \
-          \ with_tableformer=True,\n            with_easyocr=True,\n        )\n  \
-          \  elif pipeline_type == \"vlm\" and remote_model_endpoint_enabled:\n  \
-          \      # VLM pipeline with remote model endpoint: Download minimal required\
+          \ with_tableformer=True,\n            with_easyocr=False,\n        )\n \
+          \   elif pipeline_type == \"vlm\" and remote_model_endpoint_enabled:\n \
+          \       # VLM pipeline with remote model endpoint: Download minimal required\
           \ models\n        # Only models set are what lives in fabianofranz repo\n\
           \        # TODO: figure out what needs to be downloaded or removed\n   \
           \     download_models(\n            output_dir=output_path_p,\n        \
@@ -711,7 +711,7 @@ deploymentSpec:
           \        )\n    else:\n        raise ValueError(\n            f\"Invalid\
           \ pipeline_type: {pipeline_type}. Must be 'standard' or 'vlm'\"\n      \
           \  )\n\n"
-        image: quay.io/amaredia/aipcc-docling-image
+        image: quay.io/fabianofranz/docling-ubi9:2.54.0
     exec-import-pdfs:
       container:
         args:
@@ -805,7 +805,7 @@ deploymentSpec:
           \ in resp.iter_content(chunk_size=8192):\n                        if chunk:\n\
           \                            f.write(chunk)\n\n    print(\"import-test-pdfs:\
           \ done\", flush=True)\n\n"
-        image: quay.io/amaredia/aipcc-docling-image
+        image: registry.access.redhat.com/ubi9/python-311:9.6-1755074620
 pipelineInfo:
   description: Docling standard convert pipeline by the Data Processing Team
   name: data-processing-docling-standard-pipeline
@@ -972,7 +972,7 @@ root:
         isOptional: true
         parameterType: BOOLEAN
       docling_ocr_engine:
-        defaultValue: easyocr
+        defaultValue: tesseract_cli
         isOptional: true
         parameterType: STRING
       docling_pdf_backend:

kubeflow-pipelines/docling-vlm/vlm_convert_pipeline_compiled.yaml

@@ -301,7 +301,7 @@ deploymentSpec:
           \ for path in input_path_p.glob(\"*.pdf\")]\n    all_splits = [all_pdfs[i::num_splits]\
           \ for i in range(num_splits)]\n    filled_splits = list(filter(None, all_splits))\n\
           \    return filled_splits\n\n"
-        image: quay.io/amaredia/aipcc-docling-image
+        image: registry.access.redhat.com/ubi9/python-311:9.6-1755074620
     exec-docling-chunk:
       container:
         args:
@@ -415,7 +415,7 @@ deploymentSpec:
           \            f\"docling-chunk: skipped {len(skipped_files)} invalid files:\"\
           ,\n            flush=True,\n        )\n        for filename, reason in skipped_files:\n\
           \            print(f\"  - {filename}: {reason}\", flush=True)\n\n"
-        image: quay.io/amaredia/aipcc-docling-image
+        image: quay.io/fabianofranz/docling-ubi9:2.54.0
         resources:
           cpuLimit: 2.0
           cpuRequest: 0.25
@@ -547,7 +547,7 @@ deploymentSpec:
           \        result.document.save_as_markdown(\n            output_md_path,\
           \ image_mode=ImageRefMode(image_export_mode)\n        )\n\n    print(\"\
           docling-vlm-convert: done\", flush=True)\n\n"
-        image: quay.io/amaredia/aipcc-docling-image
+        image: quay.io/fabianofranz/docling-ubi9:2.54.0
         resources:
           cpuLimit: 4.0
           cpuRequest: 0.5
@@ -588,8 +588,8 @@ deploymentSpec:
           \ bool = False,\n):\n    \"\"\"\n    Download Docling models based on pipeline\
           \ type and configuration.\n\n    This unified component handles model downloading\
           \ for different pipeline types:\n    - standard : Download traditional Docling\
-          \ models (layout, tableformer, easyocr)\n    - vlm : Download Docling VLM\
-          \ models (smolvlm, smoldocling) for local inference\n             When remote_model_endpoint_enabled=True,\
+          \ models (layout, tableformer)\n    - vlm : Download Docling VLM models\
+          \ (smolvlm, smoldocling) for local inference\n             When remote_model_endpoint_enabled=True,\
           \ downloads minimal models for remote inference\n\n    Args:\n        output_path:\
           \ Path to the output directory for Docling models\n        pipeline_type:\
           \ Type of pipeline (standard, vlm)\n        remote_model_endpoint_enabled:\
@@ -601,9 +601,9 @@ deploymentSpec:
           \n    if pipeline_type == \"standard\":\n        # Standard pipeline: download\
           \ traditional models\n        download_models(\n            output_dir=output_path_p,\n\
           \            progress=True,\n            with_layout=True,\n           \
-          \ with_tableformer=True,\n            with_easyocr=True,\n        )\n  \
-          \  elif pipeline_type == \"vlm\" and remote_model_endpoint_enabled:\n  \
-          \      # VLM pipeline with remote model endpoint: Download minimal required\
+          \ with_tableformer=True,\n            with_easyocr=False,\n        )\n \
+          \   elif pipeline_type == \"vlm\" and remote_model_endpoint_enabled:\n \
+          \       # VLM pipeline with remote model endpoint: Download minimal required\
           \ models\n        # Only models set are what lives in fabianofranz repo\n\
           \        # TODO: figure out what needs to be downloaded or removed\n   \
           \     download_models(\n            output_dir=output_path_p,\n        \
@@ -624,7 +624,7 @@ deploymentSpec:
           \        )\n    else:\n        raise ValueError(\n            f\"Invalid\
           \ pipeline_type: {pipeline_type}. Must be 'standard' or 'vlm'\"\n      \
           \  )\n\n"
-        image: quay.io/amaredia/aipcc-docling-image
+        image: quay.io/fabianofranz/docling-ubi9:2.54.0
     exec-import-pdfs:
       container:
         args:
@@ -718,7 +718,7 @@ deploymentSpec:
           \ in resp.iter_content(chunk_size=8192):\n                        if chunk:\n\
           \                            f.write(chunk)\n\n    print(\"import-test-pdfs:\
           \ done\", flush=True)\n\n"
-        image: quay.io/amaredia/aipcc-docling-image
+        image: registry.access.redhat.com/ubi9/python-311:9.6-1755074620
 pipelineInfo:
   description: Docling VLM convert pipeline by the Data Processing Team
   name: data-processing-docling-vlm-pipeline