* make new OP ImageRemoveBackgroundMapper stable

data_juicer/ops/mapper/image_remove_background_mapper.py

@@ -10,6 +10,7 @@
 from ..op_fusion import LOADED_IMAGES
 
 rembg = LazyLoader('rembg', 'rembg')
+onnxruntime = LazyLoader('onnxruntime', 'onnxruntime')
 
 OP_NAME = 'image_remove_background_mapper'
 

docs/Operators.md

@@ -189,7 +189,7 @@ All the specific operators are listed below, each featured with several capabili
 | image_captioning_mapper | 🔮Multimodal 🚀GPU 🧩HF 🟢Stable | Mapper to generate samples whose captions are generated based on another model and the figure. 映射器生成样本，其标题是基于另一个模型和图生成的。 | [code](../data_juicer/ops/mapper/image_captioning_mapper.py) | [tests](../tests/ops/mapper/test_image_captioning_mapper.py) |
 | image_diffusion_mapper | 🔮Multimodal 🚀GPU 🧩HF 🟢Stable | Generate image by diffusion model. 通过扩散模型生成图像。 | [code](../data_juicer/ops/mapper/image_diffusion_mapper.py) | [tests](../tests/ops/mapper/test_image_diffusion_mapper.py) |
 | image_face_blur_mapper | 🏞Image 💻CPU 🟢Stable | Mapper to blur faces detected in images. 映射器模糊图像中检测到的人脸。 | [code](../data_juicer/ops/mapper/image_face_blur_mapper.py) | [tests](../tests/ops/mapper/test_image_face_blur_mapper.py) |
-| image_remove_background_mapper | 🏞Image 💻CPU 🟡Beta | Mapper to remove background of images. 映射器删除图像的背景。 | [code](../data_juicer/ops/mapper/image_remove_background_mapper.py) | [tests](../tests/ops/mapper/test_image_remove_background_mapper.py) |
+| image_remove_background_mapper | 🏞Image 💻CPU 🟢Stable | Mapper to remove background of images. 映射器删除图像的背景。 | [code](../data_juicer/ops/mapper/image_remove_background_mapper.py) | [tests](../tests/ops/mapper/test_image_remove_background_mapper.py) |
 | image_segment_mapper | 🏞Image 🚀GPU 🟢Stable | Perform segment-anything on images and return the bounding boxes. 对图像执行segment-任何操作并返回边界框。 | [code](../data_juicer/ops/mapper/image_segment_mapper.py) | [tests](../tests/ops/mapper/test_image_segment_mapper.py) |
 | image_tagging_mapper | 🏞Image 🚀GPU 🟢Stable | Mapper to generate image tags. 映射器生成图像标签。 | [code](../data_juicer/ops/mapper/image_tagging_mapper.py) | [tests](../tests/ops/mapper/test_image_tagging_mapper.py) |
 | mllm_mapper | 🔮Multimodal 🚀GPU 🧩HF 🟢Stable | Mapper to use MLLMs for visual question answering tasks. Mapper使用MLLMs进行视觉问答任务。 | [code](../data_juicer/ops/mapper/mllm_mapper.py) | [tests](../tests/ops/mapper/test_mllm_mapper.py) |

environments/science_requires.txt

@@ -30,3 +30,5 @@ dashscope
 openai
 ultralytics
 huggingface_hub<0.26.0
+rembg
+onnxruntime

tests/ops/mapper/test_image_remove_background_mapper.py

@@ -1,17 +1,12 @@
 import os
 import unittest
 
-import numpy as np
-
 from data_juicer.core.data import NestedDataset as Dataset
 from data_juicer.ops.mapper.image_remove_background_mapper import ImageRemoveBackgroundMapper
-from data_juicer.utils.mm_utils import load_image
-from data_juicer.utils.unittest_utils import DataJuicerTestCaseBase, SKIPPED_TESTS
+from data_juicer.utils.unittest_utils import DataJuicerTestCaseBase
+from data_juicer.utils.constant import Fields
 
 
-# Skip tests for this OP in the GitHub actions due to ?
-# These tests have been tested locally.
-@SKIPPED_TESTS.register_module()
 class ImageRemoveBackgroundMapperTest(DataJuicerTestCaseBase):
 
     data_path = os.path.join(os.path.dirname(os.path.realpath(__file__)), '..',
@@ -23,24 +18,14 @@ class ImageRemoveBackgroundMapperTest(DataJuicerTestCaseBase):
     img5_path = os.path.join(data_path, 'img5.jpg')
     img6_path = os.path.join(data_path, 'img6.jpg')
 
-
     def _run_mapper(self, op, source_list):
         dataset = Dataset.from_list(source_list)
         dataset = dataset.map(op.process)
         res_list = dataset.to_list()
-        temp_path = 'temp4test.png'
-        try:
-            from rembg import remove
-            for source, res in zip(source_list, res_list):
-                for src_path, res_path in zip(source[op.image_key], res[op.image_key]):
-                    # Compare results
-                    expected = np.array(load_image(temp_path))
-                    actual = np.array(load_image(res_path))
-                    np.testing.assert_array_equal(actual, expected)
-        finally:
-            if os.path.exists(temp_path):
-                os.remove(temp_path)
-
+        for source, res in zip(source_list, res_list):
+            for src_path, res_path in zip(source[op.image_key], res[op.image_key]):
+                self.assertNotEqual(src_path, res_path)
+            self.assertIn(Fields.source_file, res)
 
     def test_single_image(self):
         ds_list = [{
@@ -53,7 +38,6 @@ def test_single_image(self):
         op = ImageRemoveBackgroundMapper()
         self._run_mapper(op, ds_list)
 
-
     def test_multiple_images(self):
         ds_list = [{
             'images': [self.img1_path, self.img4_path]