From beaafbfaa8fdcafe7cd9fbf9b1d497710a19bc55 Mon Sep 17 00:00:00 2001
From: hoobnn <111053672+hoobnn@users.noreply.github.com>
Date: Mon, 31 Aug 2026 09:26:01 +0800
Subject: [PATCH] feat: warn on large text detector inputs

---
 .../text_detection/_large_input_warning.py    | 37 +++++++++++
 .../models/text_detection/processors.py       |  4 ++
 .../models/text_detection/test_processors.py  | 66 +++++++++++++++++++
 3 files changed, 107 insertions(+)
 create mode 100644 paddlex/inference/models/text_detection/_large_input_warning.py
 create mode 100644 tests/inference/models/text_detection/test_processors.py

diff --git a/paddlex/inference/models/text_detection/_large_input_warning.py b/paddlex/inference/models/text_detection/_large_input_warning.py
new file mode 100644
index 0000000000..ab5b487e53
--- /dev/null
+++ b/paddlex/inference/models/text_detection/_large_input_warning.py
@@ -0,0 +1,37 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+LARGE_DET_INPUT_WARN_PIXELS = 4_000_000
+
+
+class LargeDetectorInputWarner:
+    def __init__(self):
+        self._emitted = False
+
+    def warn(self, source_shape, detector_shape, warning):
+        detector_h, detector_w = detector_shape[:2]
+        if self._emitted or detector_h * detector_w <= LARGE_DET_INPUT_WARN_PIXELS:
+            return
+
+        source_h, source_w = source_shape[:2]
+        self._emitted = True
+        warning(
+            f"Text detection preprocessing produced a large detector input of "
+            f"{detector_w}x{detector_h} "
+            f"({detector_w * detector_h / 1e6:.1f} megapixels) from a "
+            f"{source_w}x{source_h} source image. This may increase memory use "
+            f"or cause an out-of-memory error. Consider lowering the detector "
+            f"input-size limits in the pipeline or model configuration."
+        )
diff --git a/paddlex/inference/models/text_detection/processors.py b/paddlex/inference/models/text_detection/processors.py
index 23a38fdd65..e8b060e728 100644
--- a/paddlex/inference/models/text_detection/processors.py
+++ b/paddlex/inference/models/text_detection/processors.py
@@ -20,6 +20,7 @@
 from ....utils import logging
 from ....utils.deps import class_requires_deps, is_dep_available
 from ...utils.benchmark import benchmark
+from ._large_input_warning import LargeDetectorInputWarner
 
 if is_dep_available("opencv-contrib-python"):
     import cv2
@@ -35,6 +36,7 @@ class DetResizeForTest:
     def __init__(self, input_shape=None, max_side_limit=4000, **kwargs):
         self.resize_type = 0
         self.keep_ratio = False
+        self._large_input_warner = LargeDetectorInputWarner()
         if input_shape is not None:
             self.input_shape = input_shape
             self.resize_type = 3
@@ -71,6 +73,8 @@ def __call__(
             img, shape = self.resize(
                 ori_img, limit_side_len, limit_type, max_side_limit
             )
+            if img is not None:
+                self._large_input_warner.warn(ori_img.shape, img.shape, logging.warning)
             resize_imgs.append(img)
             img_shapes.append(shape)
         return resize_imgs, img_shapes
diff --git a/tests/inference/models/text_detection/test_processors.py b/tests/inference/models/text_detection/test_processors.py
new file mode 100644
index 0000000000..495437e968
--- /dev/null
+++ b/tests/inference/models/text_detection/test_processors.py
@@ -0,0 +1,66 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import importlib.util
+from pathlib import Path
+from unittest.mock import Mock
+
+
+REPO_ROOT = Path(__file__).resolve().parents[4]
+spec = importlib.util.spec_from_file_location(
+    "large_input_warning",
+    REPO_ROOT / "paddlex/inference/models/text_detection/_large_input_warning.py",
+)
+large_input_warning = importlib.util.module_from_spec(spec)
+spec.loader.exec_module(large_input_warning)
+
+
+LARGE_DET_INPUT_WARN_PIXELS = large_input_warning.LARGE_DET_INPUT_WARN_PIXELS
+LargeDetectorInputWarner = large_input_warning.LargeDetectorInputWarner
+
+
+def test_large_detector_input_warning_boundary():
+    warner = LargeDetectorInputWarner()
+    warning = Mock()
+
+    warner.warn(
+        (1000, 4000, 3),
+        (1000, LARGE_DET_INPUT_WARN_PIXELS // 1000, 3),
+        warning,
+    )
+
+    warning.assert_not_called()
+
+
+def test_large_detector_input_warning_is_bounded_per_processor():
+    warner = LargeDetectorInputWarner()
+    warning = Mock()
+
+    warner.warn((3000, 2000, 3), (3000, 2000, 3), warning)
+    warner.warn((4000, 3000, 3), (4000, 3000, 3), warning)
+
+    warning.assert_called_once()
+
+
+def test_large_detector_input_warning_describes_detector_and_source_paths():
+    warner = LargeDetectorInputWarner()
+    warning = Mock()
+
+    warner.warn((3000, 4000, 3), (2048, 2048, 3), warning)
+
+    message = warning.call_args.args[0]
+    assert "2048x2048" in message
+    assert "4000x3000 source image" in message
+    assert "without downscaling" not in message
+    assert "text_det_limit" not in message