Added fast processor for llava-next-video model

src/transformers/__init__.py

@@ -1356,6 +1356,7 @@
     _import_structure["models.got_ocr2"].append("GotOcr2ImageProcessorFast")
     _import_structure["models.llava"].append("LlavaImageProcessorFast")
     _import_structure["models.llava_next"].append("LlavaNextImageProcessorFast")
+    _import_structure["models.llava_next_video"].append("LlavaNextVideoImageProcessorFast")
     _import_structure["models.llava_onevision"].append("LlavaOnevisionImageProcessorFast")
     _import_structure["models.phi4_multimodal"].append("Phi4MultimodalImageProcessorFast")
     _import_structure["models.pixtral"].append("PixtralImageProcessorFast")
@@ -6648,6 +6649,7 @@
         from .models.got_ocr2 import GotOcr2ImageProcessorFast
         from .models.llava import LlavaImageProcessorFast
         from .models.llava_next import LlavaNextImageProcessorFast
+        from .models.llava_next_video import LlavaNextVideoImageProcessorFast
         from .models.llava_onevision import LlavaOnevisionImageProcessorFast
         from .models.phi4_multimodal import Phi4MultimodalImageProcessorFast
         from .models.pixtral import PixtralImageProcessorFast

src/transformers/feature_extraction_utils.py

@@ -89,6 +89,8 @@ def __getitem__(self, item: str) -> Union[Any]:
 
     def __getattr__(self, item: str):
         try:
+            if item == "pixel_values" and "pixel_values_videos" in self.data:
+                return self.data["pixel_values_videos"]
             return self.data[item]
         except KeyError:
             raise AttributeError

src/transformers/models/auto/image_processing_auto.py

@@ -106,7 +106,7 @@
             ("levit", ("LevitImageProcessor",)),
             ("llava", ("LlavaImageProcessor", "LlavaImageProcessorFast")),
             ("llava_next", ("LlavaNextImageProcessor", "LlavaNextImageProcessorFast")),
-            ("llava_next_video", ("LlavaNextVideoImageProcessor",)),
+            ("llava_next_video", ("LlavaNextVideoImageProcessor", "LlavaNextVideoImageProcessorFast")),
             ("llava_onevision", ("LlavaOnevisionImageProcessor", "LlavaOnevisionImageProcessorFast")),
             ("mask2former", ("Mask2FormerImageProcessor",)),
             ("maskformer", ("MaskFormerImageProcessor",)),

src/transformers/models/llava_next_video/__init__.py

@@ -20,6 +20,7 @@
 if TYPE_CHECKING:
     from .configuration_llava_next_video import *
     from .image_processing_llava_next_video import *
+    from .image_processing_llava_next_video_fast import *
     from .modeling_llava_next_video import *
     from .processing_llava_next_video import *
 else:

src/transformers/models/llava_next_video/image_processing_llava_next_video_fast.py

@@ -0,0 +1,50 @@
+# coding=utf-8
+# Copyright 2025 The HuggingFace Inc. team. 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.
+"""Fast Image processor class for LLaVa-NeXT-Video."""
+
+from ...image_processing_utils_fast import BASE_IMAGE_PROCESSOR_FAST_DOCSTRING, BaseImageProcessorFast
+from ...image_utils import OPENAI_CLIP_MEAN, OPENAI_CLIP_STD, PILImageResampling
+from ...utils import add_start_docstrings
+
+
+@add_start_docstrings(
+    "Constructs a fast LlavaNextVideo image processor.",
+    BASE_IMAGE_PROCESSOR_FAST_DOCSTRING,
+)
+class LlavaNextVideoImageProcessorFast(BaseImageProcessorFast):
+    # This generated class can be used as a starting point for the fast image processor.
+    # if the image processor is only used for simple augmentations, such as resizing, center cropping, rescaling, or normalizing,
+    # only the default values should be set in the class.
+    # If the image processor requires more complex augmentations, methods from BaseImageProcessorFast can be overridden.
+    # In most cases, only the `_preprocess` method should be overridden.
+
+    # For an example of a fast image processor requiring more complex augmentations, see `LlavaNextImageProcessorFast`.
+
+    # Default values should be checked against the slow image processor
+    # None values left after checking can be removed
+    resample = PILImageResampling.BICUBIC
+    image_mean = OPENAI_CLIP_MEAN
+    image_std = OPENAI_CLIP_STD
+    size = {"shortest_edge": 224}
+    default_to_square = False
+    crop_size = {"height": 224, "width": 224}
+    do_resize = True
+    do_center_crop = True
+    do_rescale = True
+    do_normalize = True
+    do_convert_rgb = True
+    image_grid_pinpoints = [[336, 672], [672, 336], [672, 672], [1008, 336], [336, 1008]]
+
+__all__ = ["LlavaNextVideoImageProcessorFast"]

src/transformers/utils/dummy_torchvision_objects.py

@@ -86,6 +86,13 @@ def __init__(self, *args, **kwargs):
         requires_backends(self, ["torchvision"])
 
 
+class LlavaNextVideoImageProcessorFast(metaclass=DummyObject):
+    _backends = ["torchvision"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torchvision"])
+
+
 class LlavaOnevisionImageProcessorFast(metaclass=DummyObject):
     _backends = ["torchvision"]
 

tests/models/llava_next_video/test_image_processing_llava_next_video.py

@@ -19,7 +19,7 @@
 
 from transformers.image_utils import OPENAI_CLIP_MEAN, OPENAI_CLIP_STD
 from transformers.testing_utils import require_torch, require_vision
-from transformers.utils import is_torch_available, is_vision_available
+from transformers.utils import is_torch_available, is_torchvision_available, is_vision_available
 
 from ...test_image_processing_common import ImageProcessingTestMixin, prepare_image_inputs
 
@@ -32,6 +32,9 @@
 
     from transformers import LlavaNextVideoImageProcessor
 
+    if is_torchvision_available():
+        from transformers import LlavaNextVideoImageProcessorFast
+
 
 class LlavaNextVideoProcessingTester:
     def __init__(
@@ -128,6 +131,7 @@ def prepare_video_inputs(self, equal_resolution=False, numpify=False, torchify=F
 @require_vision
 class LlavaNextVideoProcessingTest(ImageProcessingTestMixin, unittest.TestCase):
     image_processing_class = LlavaNextVideoImageProcessor if is_vision_available() else None
+    fast_image_processing_class = LlavaNextVideoImageProcessorFast if is_torchvision_available() else None
 
     def setUp(self):
         super().setUp()