Refactor image loading in tests to use load_test_image helper

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
@@ -35,6 +34,7 @@
 )
 
 from ...test_configuration_common import ConfigTester
+from ...test_image_processing_common import load_coco_image
 from ...test_modeling_common import (
     TEST_EAGER_MATCHES_SDPA_INFERENCE_PARAMETERIZATION,
     ModelTesterMixin,
@@ -58,8 +58,6 @@
 
 
 if is_vision_available():
-    from PIL import Image
-
     from transformers import AutoImageProcessor, AutoProcessor
 
 
@@ -479,7 +477,7 @@ 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_coco_image("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 +511,7 @@ 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_coco_image("000000039769.jpg")
         inputs = processor(image, return_tensors="pt").to(model.device)
 
         with torch.no_grad():
@@ -544,7 +542,7 @@ 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_coco_image("000000039769.jpg")
         inputs = processor(image, return_tensors="pt").to(model.device)
 
         with torch.no_grad():

tests/models/aria/test_modeling_aria.py

@@ -16,7 +16,6 @@
 import unittest
 
 import pytest
-import requests
 
 from transformers import (
     AriaConfig,
@@ -27,7 +26,6 @@
     AutoTokenizer,
     BitsAndBytesConfig,
     is_torch_available,
-    is_vision_available,
 )
 from transformers.models.idefics3 import Idefics3VisionConfig
 from transformers.testing_utils import (
@@ -43,16 +41,14 @@
 
 from ...generation.test_utils import GenerationTesterMixin
 from ...test_configuration_common import ConfigTester
+from ...test_image_processing_common import load_coco_image, load_test_image
 from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor
 
 
 if is_torch_available():
     import torch
 
 
-if is_vision_available():
-    from PIL import Image
-
 # Used to be https://aria-vl.github.io/static/images/view.jpg but it was removed, llava-vl has the same image
 IMAGE_OF_VIEW_URL = "https://llava-vl.github.io/static/images/view.jpg"
 
@@ -246,7 +242,7 @@ def test_small_model_integration_test(self):
         )
 
         prompt = "<|img|>\nUSER: What are the things I should be cautious about when I visit this place?\nASSISTANT:"
-        raw_image = Image.open(requests.get(IMAGE_OF_VIEW_URL, stream=True).raw)
+        raw_image = load_test_image(IMAGE_OF_VIEW_URL)
         inputs = self.processor(images=raw_image, text=prompt, return_tensors="pt").to(model.device, model.dtype)
 
         non_img_tokens = [
@@ -286,7 +282,7 @@ def test_small_model_integration_test_llama_single(self):
         processor = AutoProcessor.from_pretrained(model_id)
 
         prompt = "USER: <|img|>\nWhat are the things I should be cautious about when I visit this place? ASSISTANT:"
-        raw_image = Image.open(requests.get(IMAGE_OF_VIEW_URL, stream=True).raw)
+        raw_image = load_test_image(IMAGE_OF_VIEW_URL)
         inputs = processor(images=raw_image, text=prompt, return_tensors="pt").to(model.device, model.dtype)
 
         output = model.generate(**inputs, max_new_tokens=90, do_sample=False)
@@ -320,8 +316,8 @@ def test_small_model_integration_test_llama_batched(self):
             "USER: <|img|>\nWhat are the things I should be cautious about when I visit this place? What should I bring with me? ASSISTANT:",
             "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)
+        image1 = load_test_image(IMAGE_OF_VIEW_URL)
+        image2 = load_coco_image("000000039769.jpg")
 
         inputs = processor(images=[image1, image2], text=prompts, return_tensors="pt", padding=True).to(
             model.device, model.dtype
@@ -358,8 +354,8 @@ def test_small_model_integration_test_batch(self):
             "USER: <|img|>\nWhat are the things I should be cautious about when I visit this place? What should I bring with me?\nASSISTANT:",
             "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)
+        image1 = load_test_image(IMAGE_OF_VIEW_URL)
+        image2 = load_coco_image("000000039769.jpg")
 
         inputs = self.processor(images=[image1, image2], text=prompts, return_tensors="pt", padding=True).to(
             model.device, model.dtype
@@ -398,8 +394,8 @@ def test_small_model_integration_test_llama_batched_regression(self):
             "USER: <|img|>\nWhat are the things I should be cautious about when I visit this place? What should I bring with me?\nASSISTANT:",
             "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)
+        image1 = load_test_image(IMAGE_OF_VIEW_URL)
+        image2 = load_coco_image("000000039769.jpg")
 
         inputs = processor(images=[image1, image2, image1], text=prompts, return_tensors="pt", padding=True)
         inputs = inputs.to(model.device, model.dtype)
@@ -431,8 +427,8 @@ def test_batched_generation(self):
         prompt3 = "<image>\nUSER: Describe the image.\nASSISTANT:"
         url1 = "https://images.unsplash.com/photo-1552053831-71594a27632d?q=80&w=3062&auto=format&fit=crop&ixlib=rb-4.0.3&ixid=M3wxMjA3fDB8MHxwaG90by1wYWdlfHx8fGVufDB8fHx8fA%3D%3D"
         url2 = "https://images.unsplash.com/photo-1617258683320-61900b281ced?q=80&w=3087&auto=format&fit=crop&ixlib=rb-4.0.3&ixid=M3wxMjA3fDB8MHxwaG90by1wYWdlfHx8fGVufDB8fHx8fA%3D%3D"
-        image1 = Image.open(requests.get(url1, stream=True).raw)
-        image2 = Image.open(requests.get(url2, stream=True).raw)
+        image1 = load_test_image(url1)
+        image2 = load_test_image(url2)
 
         # Create inputs
         messages = [

tests/models/deepseek_vl_hybrid/test_image_processing_deepseek_vl_hybrid.py

@@ -19,7 +19,11 @@
 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_image_processing_common import (
+    ImageProcessingTestMixin,
+    load_coco_image,
+    prepare_image_inputs,
+)
 
 
 if is_torch_available():
@@ -203,15 +207,7 @@ 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_coco_image("000000039769.jpg")
 
         encodings = {}
         for backend_name, image_processing_class in self.image_processing_classes.items():

tests/models/eomt/test_modeling_eomt.py

@@ -15,13 +15,12 @@
 
 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_image_processing_common import load_coco_image
 from ...test_modeling_common import ModelTesterMixin, floats_tensor
 from ...test_pipeline_mixin import PipelineTesterMixin
 
@@ -30,10 +29,6 @@
     import torch
 
 
-if is_vision_available():
-    from PIL import Image
-
-
 class EomtForUniversalSegmentationTester:
     def __init__(
         self,
@@ -175,7 +170,7 @@ 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_coco_image("000000039769.jpg")
 
         inputs = processor(images=image, return_tensors="pt").to(model.device)
 
@@ -218,7 +213,7 @@ 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_coco_image("000000039769.jpg")
 
         inputs = processor(images=image, return_tensors="pt").to(model.device)
 
@@ -234,7 +229,7 @@ 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_coco_image("000000039769.jpg")
 
         inputs = processor(images=image, return_tensors="pt").to(model.device)
 
@@ -271,7 +266,7 @@ 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_coco_image("000000039769.jpg")
 
         inputs = processor(images=image, return_tensors="pt").to(model.device)
 
@@ -320,7 +315,7 @@ 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_coco_image("000000039769.jpg")
 
         inputs = processor(images=image, return_tensors="pt").to(model.device)
 
@@ -366,7 +361,7 @@ 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_coco_image("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,
@@ -26,9 +24,10 @@
     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_image_processing_common import load_coco_image
 from ...test_modeling_common import ModelTesterMixin, _config_zero_init, floats_tensor
 from ...test_pipeline_mixin import PipelineTesterMixin
 
@@ -37,10 +36,6 @@
     import torch
 
 
-if is_vision_available():
-    from PIL import Image
-
-
 class EomtDinov3ForUniversalSegmentationTester:
     def __init__(
         self,
@@ -220,7 +215,7 @@ 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_coco_image("000000039769.jpg")
 
         inputs = processor(images=image, return_tensors="pt").to(model.device)
 
@@ -262,7 +257,7 @@ 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_coco_image("000000039769.jpg")
 
         inputs = processor(images=image, return_tensors="pt").to(model.device)
 
@@ -319,7 +314,7 @@ 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_coco_image("000000039769.jpg")
 
         inputs = processor(images=image, return_tensors="pt").to(model.device)
 
@@ -375,7 +370,7 @@ 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_coco_image("000000039769.jpg")
 
         inputs = processor(images=image, return_tensors="pt").to(model.device)
 
@@ -421,7 +416,7 @@ 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_coco_image("000000039769.jpg")
 
         inputs = processor(images=image, return_tensors="pt").to(model.device)
 
@@ -463,7 +458,7 @@ 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_coco_image("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

@@ -16,15 +16,12 @@
 import copy
 import unittest
 
-import requests
-
 from transformers import (
     AutoProcessor,
     FastVlmConfig,
     FastVlmForConditionalGeneration,
     FastVlmModel,
     is_torch_available,
-    is_vision_available,
 )
 from transformers.testing_utils import (
     Expectations,
@@ -38,17 +35,14 @@
 
 from ...generation.test_utils import GenerationTesterMixin
 from ...test_configuration_common import ConfigTester
+from ...test_image_processing_common import load_coco_image, load_test_image
 from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor
 
 
 if is_torch_available():
     import torch
 
 
-if is_vision_available():
-    from PIL import Image
-
-
 class FastVlmVisionText2TextModelTester:
     def __init__(
         self,
@@ -257,7 +251,7 @@ def test_small_model_integration_test(self):
 
         prompt = "user\n<image>\nWhat are the things I should be cautious about when I visit this place?\nassistant"
         image_file = "https://llava-vl.github.io/static/images/view.jpg"
-        raw_image = Image.open(requests.get(image_file, stream=True).raw)
+        raw_image = load_test_image(image_file)
         inputs = self.processor(images=raw_image, text=prompt, return_tensors="pt").to(torch_device, dtype=model.dtype)
 
         output = model.generate(**inputs, max_new_tokens=20)
@@ -281,8 +275,8 @@ def test_small_model_integration_test_batch(self):
             "user\n<image>\nWhat are the things I should be cautious about when I visit this place? What should I bring with me?\nassistant",
             "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)
+        image1 = load_test_image("https://llava-vl.github.io/static/images/view.jpg")
+        image2 = load_coco_image("000000039769.jpg")
 
         self.processor.tokenizer.padding_side = "left"
         inputs = self.processor(images=[image1, image2], text=prompts, return_tensors="pt", padding=True).to(