nv-ingest-api 2025.7.13.dev20250713__tar.gz → 2025.7.15.dev20250715__tar.gz

{nv_ingest_api-2025.7.13.dev20250713/src/nv_ingest_api.egg-info → nv_ingest_api-2025.7.15.dev20250715}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nv-ingest-api
-Version: 2025.7.13.dev20250713
+Version: 2025.7.15.dev20250715
 Summary: Python module with core document ingestion functions.
 Author-email: Jeremy Dyer <jdyer@nvidia.com>
 License:                                  Apache License
@@ -217,6 +217,7 @@ Requires-Dist: backoff==2.2.1
 Requires-Dist: pandas>=2.0
 Requires-Dist: pydantic>2.0.0
 Requires-Dist: pydantic-settings>2.0.0
+Requires-Dist: tritonclient
 Dynamic: license-file
 
 # nv-ingest-api

{nv_ingest_api-2025.7.13.dev20250713 → nv_ingest_api-2025.7.15.dev20250715}/pyproject.toml

@@ -21,6 +21,7 @@ dependencies = [
     "pandas>=2.0",
     "pydantic>2.0.0",
     "pydantic-settings>2.0.0",
+    "tritonclient",
 ]
 
 [project.urls]

{nv_ingest_api-2025.7.13.dev20250713 → nv_ingest_api-2025.7.15.dev20250715}/src/nv_ingest_api/internal/extract/pdf/engines/nemoretriever.py

@@ -40,6 +40,7 @@ from nv_ingest_api.internal.schemas.meta.metadata_schema import validate_metadat
 from nv_ingest_api.internal.primitives.nim.model_interface.yolox import (
     YOLOX_PAGE_IMAGE_PREPROC_WIDTH,
     YOLOX_PAGE_IMAGE_PREPROC_HEIGHT,
+    YOLOX_PAGE_IMAGE_FORMAT,
 )
 from nv_ingest_api.internal.schemas.extract.extract_pdf_schema import NemoRetrieverParseConfigSchema
 from nv_ingest_api.util.metadata.aggregators import (
@@ -355,7 +356,7 @@ def nemoretriever_parse_extractor(
                 img_numpy = crop_image(page_image, transformed_bbox)
 
                 if img_numpy is not None:
-                    base64_img = numpy_to_base64(img_numpy)
+                    base64_img = numpy_to_base64(img_numpy, format=YOLOX_PAGE_IMAGE_FORMAT)
                     image = Base64Image(
                         image=base64_img,
                         bbox=transformed_bbox,

{nv_ingest_api-2025.7.13.dev20250713 → nv_ingest_api-2025.7.15.dev20250715}/src/nv_ingest_api/internal/extract/pdf/engines/pdfium.py

@@ -28,6 +28,7 @@ from nv_ingest_api.internal.primitives.nim.default_values import YOLOX_MAX_BATCH
 from nv_ingest_api.internal.primitives.nim.model_interface.yolox import (
     YOLOX_PAGE_IMAGE_PREPROC_WIDTH,
     YOLOX_PAGE_IMAGE_PREPROC_HEIGHT,
+    YOLOX_PAGE_IMAGE_FORMAT,
     get_yolox_model_name,
     YoloxPageElementsModelInterface,
 )
@@ -186,7 +187,7 @@ def _extract_page_element_images(
             if cropped is None:
                 continue
 
-            base64_img = numpy_to_base64(cropped)
+            base64_img = numpy_to_base64(cropped, format=YOLOX_PAGE_IMAGE_FORMAT)
 
             bbox_in_orig_coord = (
                 int(w1) - pad_width,

{nv_ingest_api-2025.7.13.dev20250713 → nv_ingest_api-2025.7.15.dev20250715}/src/nv_ingest_api/internal/primitives/nim/model_interface/nemoretriever_parse.py

@@ -120,6 +120,7 @@ class NemoRetrieverParseModelInterface(ModelInterface):
             logger.debug("Formatting input for HTTP NemoRetrieverParse model")
             # Prepare payload for HTTP request
 
+            ## TODO: Ask @Edward Kim if we want to switch to JPEG/PNG here
             if "images" in data:
                 base64_list = [numpy_to_base64(img) for img in data["images"]]
             else:

{nv_ingest_api-2025.7.13.dev20250713 → nv_ingest_api-2025.7.15.dev20250715}/src/nv_ingest_api/internal/primitives/nim/model_interface/yolox.py

@@ -2,9 +2,7 @@
 # All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 
-
-import base64
-import io
+import os
 import logging
 import warnings
 from math import log
@@ -20,11 +18,11 @@ import packaging
 import pandas as pd
 import torch
 import torchvision
-from PIL import Image
 
 from nv_ingest_api.internal.primitives.nim import ModelInterface
 from nv_ingest_api.internal.primitives.nim.model_interface.helpers import get_model_name
 from nv_ingest_api.util.image_processing import scale_image_to_encoding_size
+from nv_ingest_api.util.image_processing.transforms import numpy_to_base64
 
 logger = logging.getLogger(__name__)
 
@@ -35,6 +33,7 @@ YOLOX_PAGE_MIN_SCORE = 0.1
 YOLOX_PAGE_NIM_MAX_IMAGE_SIZE = 512_000
 YOLOX_PAGE_IMAGE_PREPROC_HEIGHT = 1024
 YOLOX_PAGE_IMAGE_PREPROC_WIDTH = 1024
+YOLOX_PAGE_IMAGE_FORMAT = os.getenv("YOLOX_PAGE_IMAGE_FORMAT", "PNG")
 
 # yolox-page-elements-v1 contants
 YOLOX_PAGE_V1_NUM_CLASSES = 4
@@ -239,15 +238,11 @@ class YoloxModelInterfaceBase(ModelInterface):
                 # Convert to uint8 if needed.
                 if image.dtype != np.uint8:
                     image = (image * 255).astype(np.uint8)
-                # Convert the numpy array to a PIL Image.
-                image_pil = Image.fromarray(image)
-                original_size = image_pil.size
-
-                # Save the image to a buffer and encode to base64.
-                buffered = io.BytesIO()
-                image_pil.save(buffered, format="PNG")
-                image_b64 = base64.b64encode(buffered.getvalue()).decode("utf-8")
 
+                # Get original size directly from numpy array (width, height)
+                original_size = (image.shape[1], image.shape[0])
+                # Convert numpy array directly to base64 using OpenCV
+                image_b64 = numpy_to_base64(image, format=YOLOX_PAGE_IMAGE_FORMAT)
                 # Scale the image if necessary.
                 scaled_image_b64, new_size = scale_image_to_encoding_size(
                     image_b64, max_base64_size=self.nim_max_image_size