fix: resolve integration test failures by migrating COCO image loading to HF Hub fixtures

tests/models/aimv2/test_modeling_aimv2.py

@@ -18,7 +18,6 @@
 import unittest
 
 import numpy as np
-import requests
 from parameterized import parameterized
 
 from transformers import Aimv2Config, Aimv2TextConfig, Aimv2VisionConfig
@@ -44,6 +43,7 @@
     random_attention_mask,
 )
 from ...test_pipeline_mixin import PipelineTesterMixin
+from ...test_processing_common import load_test_image
 
 
 if is_torch_available():
@@ -58,8 +58,6 @@
 
 
 if is_vision_available():
-    from PIL import Image
-
     from transformers import AutoImageProcessor, AutoProcessor
 
 
@@ -479,7 +477,9 @@ def test_inference(self):
         model = Aimv2Model.from_pretrained(model_name, device_map=torch_device)
         processor = AutoProcessor.from_pretrained(model_name)
 
-        image = Image.open(requests.get("http://images.cocodataset.org/val2017/000000039769.jpg", stream=True).raw)
+        image = load_test_image(
+            "https://huggingface.co/datasets/hf-internal-testing/fixtures-coco/resolve/main/val2017/000000039769.jpg"
+        )
         inputs = processor(
             text=["a photo of a cat", "a photo of a dog"], images=image, padding=True, return_tensors="pt"
         ).to(model.device)
@@ -513,7 +513,9 @@ def test_inference(self):
         model = Aimv2VisionModel.from_pretrained(model_name, device_map=torch_device)
         processor = AutoImageProcessor.from_pretrained(model_name)
 
-        image = Image.open(requests.get("http://images.cocodataset.org/val2017/000000039769.jpg", stream=True).raw)
+        image = load_test_image(
+            "https://huggingface.co/datasets/hf-internal-testing/fixtures-coco/resolve/main/val2017/000000039769.jpg"
+        )
         inputs = processor(image, return_tensors="pt").to(model.device)
 
         with torch.no_grad():
@@ -544,7 +546,9 @@ def test_inference_for_native_resolution(self):
         model = Aimv2VisionModel.from_pretrained(model_name, device_map="auto")
         processor = AutoImageProcessor.from_pretrained(model_name)
 
-        image = Image.open(requests.get("http://images.cocodataset.org/val2017/000000039769.jpg", stream=True).raw)
+        image = load_test_image(
+            "https://huggingface.co/datasets/hf-internal-testing/fixtures-coco/resolve/main/val2017/000000039769.jpg"
+        )
         inputs = processor(image, return_tensors="pt").to(model.device)
 
         with torch.no_grad():

tests/models/aria/test_modeling_aria.py

@@ -44,6 +44,7 @@
 from ...generation.test_utils import GenerationTesterMixin
 from ...test_configuration_common import ConfigTester
 from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor
+from ...test_processing_common import load_test_image
 
 
 if is_torch_available():
@@ -321,7 +322,9 @@ def test_small_model_integration_test_llama_batched(self):
             "USER: <|img|>\nWhat is this? ASSISTANT:",
         ]
         image1 = Image.open(requests.get(IMAGE_OF_VIEW_URL, stream=True).raw)
-        image2 = Image.open(requests.get("http://images.cocodataset.org/val2017/000000039769.jpg", stream=True).raw)
+        image2 = load_test_image(
+            "https://huggingface.co/datasets/hf-internal-testing/fixtures-coco/resolve/main/val2017/000000039769.jpg"
+        )
 
         inputs = processor(images=[image1, image2], text=prompts, return_tensors="pt", padding=True).to(
             model.device, model.dtype
@@ -359,7 +362,9 @@ def test_small_model_integration_test_batch(self):
             "USER: <|img|>\nWhat is this?\nASSISTANT:",
         ]
         image1 = Image.open(requests.get(IMAGE_OF_VIEW_URL, stream=True).raw)
-        image2 = Image.open(requests.get("http://images.cocodataset.org/val2017/000000039769.jpg", stream=True).raw)
+        image2 = load_test_image(
+            "https://huggingface.co/datasets/hf-internal-testing/fixtures-coco/resolve/main/val2017/000000039769.jpg"
+        )
 
         inputs = self.processor(images=[image1, image2], text=prompts, return_tensors="pt", padding=True).to(
             model.device, model.dtype
@@ -399,7 +404,9 @@ def test_small_model_integration_test_llama_batched_regression(self):
             "USER: <|img|>\nWhat is this?\nASSISTANT: Two cats lying on a bed!\nUSER: <|img|>\nAnd this?\nASSISTANT:",
         ]
         image1 = Image.open(requests.get(IMAGE_OF_VIEW_URL, stream=True).raw)
-        image2 = Image.open(requests.get("http://images.cocodataset.org/val2017/000000039769.jpg", stream=True).raw)
+        image2 = load_test_image(
+            "https://huggingface.co/datasets/hf-internal-testing/fixtures-coco/resolve/main/val2017/000000039769.jpg"
+        )
 
         inputs = processor(images=[image1, image2, image1], text=prompts, return_tensors="pt", padding=True)
         inputs = inputs.to(model.device, model.dtype)

tests/models/deepseek_vl_hybrid/test_image_processing_deepseek_vl_hybrid.py

@@ -20,6 +20,7 @@
 from transformers.utils import is_torch_available, is_vision_available
 
 from ...test_image_processing_common import ImageProcessingTestMixin, prepare_image_inputs
+from ...test_processing_common import load_test_image
 
 
 if is_torch_available():
@@ -203,14 +204,8 @@ def test_backends_equivalence(self):
         if len(self.image_processing_classes) < 2:
             self.skipTest(reason="Skipping backends equivalence test as there are less than 2 backends")
 
-        import io
-
-        import httpx
-
-        dummy_image = Image.open(
-            io.BytesIO(
-                httpx.get("http://images.cocodataset.org/val2017/000000039769.jpg", follow_redirects=True).content
-            )
+        dummy_image = load_test_image(
+            "https://huggingface.co/datasets/hf-internal-testing/fixtures-coco/resolve/main/val2017/000000039769.jpg"
         )
 
         encodings = {}

tests/models/eomt/test_modeling_eomt.py

@@ -15,25 +15,20 @@
 
 import unittest
 
-import requests
-
 from transformers import AutoImageProcessor, EomtConfig, EomtForUniversalSegmentation, pipeline
 from transformers.testing_utils import require_torch, require_torch_accelerator, require_torch_fp16, slow, torch_device
-from transformers.utils import is_torch_available, is_vision_available
+from transformers.utils import is_torch_available
 
 from ...test_configuration_common import ConfigTester
 from ...test_modeling_common import ModelTesterMixin, floats_tensor
 from ...test_pipeline_mixin import PipelineTesterMixin
+from ...test_processing_common import load_test_image
 
 
 if is_torch_available():
     import torch
 
 
-if is_vision_available():
-    from PIL import Image
-
-
 class EomtForUniversalSegmentationTester:
     def __init__(
         self,
@@ -175,7 +170,9 @@ def test_inference(self):
         model = EomtForUniversalSegmentation.from_pretrained(self.model_id, device_map="auto")
         processor = AutoImageProcessor.from_pretrained(self.model_id)
 
-        image = Image.open(requests.get("http://images.cocodataset.org/val2017/000000039769.jpg", stream=True).raw)
+        image = load_test_image(
+            "https://huggingface.co/datasets/hf-internal-testing/fixtures-coco/resolve/main/val2017/000000039769.jpg"
+        )
 
         inputs = processor(images=image, return_tensors="pt").to(model.device)
 
@@ -218,7 +215,9 @@ def test_inference_fp16(self):
         model = EomtForUniversalSegmentation.from_pretrained(self.model_id, dtype=torch.float16, device_map="auto")
         processor = AutoImageProcessor.from_pretrained(self.model_id)
 
-        image = Image.open(requests.get("http://images.cocodataset.org/val2017/000000039769.jpg", stream=True).raw)
+        image = load_test_image(
+            "https://huggingface.co/datasets/hf-internal-testing/fixtures-coco/resolve/main/val2017/000000039769.jpg"
+        )
 
         inputs = processor(images=image, return_tensors="pt").to(model.device)
 
@@ -234,7 +233,9 @@ def test_semantic_segmentation_inference(self):
         model = EomtForUniversalSegmentation.from_pretrained(model_id, device_map="auto")
         processor = AutoImageProcessor.from_pretrained(model_id)
 
-        image = Image.open(requests.get("http://images.cocodataset.org/val2017/000000039769.jpg", stream=True).raw)
+        image = load_test_image(
+            "https://huggingface.co/datasets/hf-internal-testing/fixtures-coco/resolve/main/val2017/000000039769.jpg"
+        )
 
         inputs = processor(images=image, return_tensors="pt").to(model.device)
 
@@ -271,7 +272,9 @@ def test_panoptic_segmentation_inference(self):
         model = EomtForUniversalSegmentation.from_pretrained(self.model_id, device_map="auto")
         processor = AutoImageProcessor.from_pretrained(self.model_id)
 
-        image = Image.open(requests.get("http://images.cocodataset.org/val2017/000000039769.jpg", stream=True).raw)
+        image = load_test_image(
+            "https://huggingface.co/datasets/hf-internal-testing/fixtures-coco/resolve/main/val2017/000000039769.jpg"
+        )
 
         inputs = processor(images=image, return_tensors="pt").to(model.device)
 
@@ -320,7 +323,9 @@ def test_instance_segmentation_inference(self):
         model = EomtForUniversalSegmentation.from_pretrained(model_id, device_map="auto")
         processor = AutoImageProcessor.from_pretrained(model_id)
 
-        image = Image.open(requests.get("http://images.cocodataset.org/val2017/000000039769.jpg", stream=True).raw)
+        image = load_test_image(
+            "https://huggingface.co/datasets/hf-internal-testing/fixtures-coco/resolve/main/val2017/000000039769.jpg"
+        )
 
         inputs = processor(images=image, return_tensors="pt").to(model.device)
 
@@ -366,7 +371,9 @@ def test_instance_segmentation_inference(self):
 
     @slow
     def test_segmentation_pipeline(self):
-        image = Image.open(requests.get("http://images.cocodataset.org/val2017/000000039769.jpg", stream=True).raw)
+        image = load_test_image(
+            "https://huggingface.co/datasets/hf-internal-testing/fixtures-coco/resolve/main/val2017/000000039769.jpg"
+        )
 
         pipe = pipeline(model=self.model_id, subtask="panoptic", device=torch_device)
         output = pipe(image)

tests/models/eomt_dinov3/test_modeling_eomt_dinov3.py

@@ -15,8 +15,6 @@
 
 import unittest
 
-import requests
-
 from transformers import AutoImageProcessor, EomtDinov3Config, EomtDinov3ForUniversalSegmentation, pipeline
 from transformers.testing_utils import (
     Expectations,
@@ -26,21 +24,18 @@
     slow,
     torch_device,
 )
-from transformers.utils import is_torch_available, is_vision_available
+from transformers.utils import is_torch_available
 
 from ...test_configuration_common import ConfigTester
 from ...test_modeling_common import ModelTesterMixin, _config_zero_init, floats_tensor
 from ...test_pipeline_mixin import PipelineTesterMixin
+from ...test_processing_common import load_test_image
 
 
 if is_torch_available():
     import torch
 
 
-if is_vision_available():
-    from PIL import Image
-
-
 class EomtDinov3ForUniversalSegmentationTester:
     def __init__(
         self,
@@ -220,7 +215,9 @@ def test_inference(self):
         model = EomtDinov3ForUniversalSegmentation.from_pretrained(self.model_id, device_map="auto")
         processor = AutoImageProcessor.from_pretrained(self.model_id)
 
-        image = Image.open(requests.get("http://images.cocodataset.org/val2017/000000039769.jpg", stream=True).raw)
+        image = load_test_image(
+            "https://huggingface.co/datasets/hf-internal-testing/fixtures-coco/resolve/main/val2017/000000039769.jpg"
+        )
 
         inputs = processor(images=image, return_tensors="pt").to(model.device)
 
@@ -262,7 +259,9 @@ def test_inference_bf16(self):
         )
         processor = AutoImageProcessor.from_pretrained(self.model_id)
 
-        image = Image.open(requests.get("http://images.cocodataset.org/val2017/000000039769.jpg", stream=True).raw)
+        image = load_test_image(
+            "https://huggingface.co/datasets/hf-internal-testing/fixtures-coco/resolve/main/val2017/000000039769.jpg"
+        )
 
         inputs = processor(images=image, return_tensors="pt").to(model.device)
 
@@ -319,7 +318,9 @@ def test_semantic_segmentation_inference(self):
         model = EomtDinov3ForUniversalSegmentation.from_pretrained(model_id, device_map="auto")
         processor = AutoImageProcessor.from_pretrained(model_id)
 
-        image = Image.open(requests.get("http://images.cocodataset.org/val2017/000000039769.jpg", stream=True).raw)
+        image = load_test_image(
+            "https://huggingface.co/datasets/hf-internal-testing/fixtures-coco/resolve/main/val2017/000000039769.jpg"
+        )
 
         inputs = processor(images=image, return_tensors="pt").to(model.device)
 
@@ -375,7 +376,9 @@ def test_panoptic_segmentation_inference(self):
         model = EomtDinov3ForUniversalSegmentation.from_pretrained(self.model_id, device_map="auto")
         processor = AutoImageProcessor.from_pretrained(self.model_id)
 
-        image = Image.open(requests.get("http://images.cocodataset.org/val2017/000000039769.jpg", stream=True).raw)
+        image = load_test_image(
+            "https://huggingface.co/datasets/hf-internal-testing/fixtures-coco/resolve/main/val2017/000000039769.jpg"
+        )
 
         inputs = processor(images=image, return_tensors="pt").to(model.device)
 
@@ -421,7 +424,9 @@ def test_instance_segmentation_inference(self):
         model = EomtDinov3ForUniversalSegmentation.from_pretrained(model_id, device_map="auto")
         processor = AutoImageProcessor.from_pretrained(model_id)
 
-        image = Image.open(requests.get("http://images.cocodataset.org/val2017/000000039769.jpg", stream=True).raw)
+        image = load_test_image(
+            "https://huggingface.co/datasets/hf-internal-testing/fixtures-coco/resolve/main/val2017/000000039769.jpg"
+        )
 
         inputs = processor(images=image, return_tensors="pt").to(model.device)
 
@@ -463,7 +468,9 @@ def test_instance_segmentation_inference(self):
             self.assertTrue(0.0 <= info["score"] <= 1.0)
 
     def test_segmentation_pipeline(self):
-        image = Image.open(requests.get("http://images.cocodataset.org/val2017/000000039769.jpg", stream=True).raw)
+        image = load_test_image(
+            "https://huggingface.co/datasets/hf-internal-testing/fixtures-coco/resolve/main/val2017/000000039769.jpg"
+        )
 
         model = EomtDinov3ForUniversalSegmentation.from_pretrained(self.model_id, device_map="auto")
         processor = AutoImageProcessor.from_pretrained(self.model_id)

tests/models/fast_vlm/test_modeling_fast_vlm.py

@@ -39,6 +39,7 @@
 from ...generation.test_utils import GenerationTesterMixin
 from ...test_configuration_common import ConfigTester
 from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor
+from ...test_processing_common import load_test_image
 
 
 if is_torch_available():
@@ -282,7 +283,9 @@ def test_small_model_integration_test_batch(self):
             "user\n<image>\nWhat is this?\nassistant",
         ]
         image1 = Image.open(requests.get("https://llava-vl.github.io/static/images/view.jpg", stream=True).raw)
-        image2 = Image.open(requests.get("http://images.cocodataset.org/val2017/000000039769.jpg", stream=True).raw)
+        image2 = load_test_image(
+            "https://huggingface.co/datasets/hf-internal-testing/fixtures-coco/resolve/main/val2017/000000039769.jpg"
+        )
 
         self.processor.tokenizer.padding_side = "left"
         inputs = self.processor(images=[image1, image2], text=prompts, return_tensors="pt", padding=True).to(

tests/models/flava/test_image_processing_flava.py

@@ -12,18 +12,16 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import io
 import random
 import unittest
 
-import httpx
 import numpy as np
-from PIL import Image
 
 from transformers.testing_utils import require_torch, require_vision
 from transformers.utils import is_torch_available, is_vision_available
 
 from ...test_image_processing_common import ImageProcessingTestMixin, prepare_image_inputs
+from ...test_processing_common import load_test_image
 
 
 if is_torch_available():
@@ -404,10 +402,8 @@ def test_slow_fast_equivalence(self):
         if len(self.image_processing_classes) < 2:
             self.skipTest(reason="Skipping backends equivalence test as there are less than 2 backends")
 
-        dummy_image = Image.open(
-            io.BytesIO(
-                httpx.get("http://images.cocodataset.org/val2017/000000039769.jpg", follow_redirects=True).content
-            )
+        dummy_image = load_test_image(
+            "https://huggingface.co/datasets/hf-internal-testing/fixtures-coco/resolve/main/val2017/000000039769.jpg"
         )
 
         # Create processors for each backend

tests/models/fuyu/test_image_processing_fuyu.py

@@ -1,7 +1,5 @@
-import io
 import unittest
 
-import httpx
 import numpy as np
 import pytest
 
@@ -17,6 +15,7 @@
 from transformers.utils import is_torch_available, is_vision_available
 
 from ...test_image_processing_common import ImageProcessingTestMixin
+from ...test_processing_common import load_test_image
 
 
 if is_torch_available() and is_vision_available():
@@ -180,10 +179,8 @@ def test_backends_equivalence(self):
         if len(self.image_processing_classes) < 2:
             self.skipTest(reason="Skipping backends equivalence test as there are less than 2 backends")
 
-        dummy_image = Image.open(
-            io.BytesIO(
-                httpx.get("http://images.cocodataset.org/val2017/000000039769.jpg", follow_redirects=True).content
-            )
+        dummy_image = load_test_image(
+            "https://huggingface.co/datasets/hf-internal-testing/fixtures-coco/resolve/main/val2017/000000039769.jpg"
         )
 
         encodings = {}