Fast image processor

docs/source/en/main_classes/image_processor.md

@@ -32,3 +32,8 @@ An image processor is in charge of preparing input features for vision models an
 ## BaseImageProcessor
 
 [[autodoc]] image_processing_utils.BaseImageProcessor
+
+
+## BaseImageProcessorFast
+
+[[autodoc]] image_processing_utils_fast.BaseImageProcessorFast

docs/source/en/model_doc/vit.md

@@ -62,7 +62,7 @@ Following the original Vision Transformer, some follow-up works have been made:
 This model was contributed by [nielsr](https://huggingface.co/nielsr). The original code (written in JAX) can be
 found [here](https://github.com/google-research/vision_transformer).
 
-Note that we converted the weights from Ross Wightman's [timm library](https://github.com/rwightman/pytorch-image-models), 
+Note that we converted the weights from Ross Wightman's [timm library](https://github.com/rwightman/pytorch-image-models),
 who already converted the weights from JAX to PyTorch. Credits go to him!
 
 ## Usage tips
@@ -158,6 +158,11 @@ A list of official Hugging Face and community (indicated by 🌎) resources to h
 [[autodoc]] ViTImageProcessor
     - preprocess
 
+## ViTImageProcessorFast
+
+[[autodoc]] ViTImageProcessorFast
+    - preprocess
+
 <frameworkcontent>
 <pt>
 

examples/pytorch/_tests_requirements.txt

@@ -29,3 +29,4 @@ timm
 albumentations >= 1.4.5
 torchmetrics
 pycocotools
+Pillow>=10.0.1,<=15.0

src/transformers/__init__.py

@@ -1104,7 +1104,8 @@
         name for name in dir(dummy_vision_objects) if not name.startswith("_")
     ]
 else:
-    _import_structure["image_processing_utils"] = ["ImageProcessingMixin"]
+    _import_structure["image_processing_base"] = ["ImageProcessingMixin"]
+    _import_structure["image_processing_utils"] = ["BaseImageProcessor"]
     _import_structure["image_utils"] = ["ImageFeatureExtractionMixin"]
     _import_structure["models.beit"].extend(["BeitFeatureExtractor", "BeitImageProcessor"])
     _import_structure["models.bit"].extend(["BitImageProcessor"])
@@ -1167,6 +1168,18 @@
     _import_structure["models.vivit"].append("VivitImageProcessor")
     _import_structure["models.yolos"].extend(["YolosFeatureExtractor", "YolosImageProcessor"])
 
+try:
+    if not is_torchvision_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    from .utils import dummy_torchvision_objects
+
+    _import_structure["utils.dummy_torchvision_objects"] = [
+        name for name in dir(dummy_torchvision_objects) if not name.startswith("_")
+    ]
+else:
+    _import_structure["image_processing_utils_fast"] = ["BaseImageProcessorFast"]
+    _import_structure["models.vit"].append("ViTImageProcessorFast")
 
 # PyTorch-backed objects
 try:
@@ -5703,7 +5716,8 @@
     except OptionalDependencyNotAvailable:
         from .utils.dummy_vision_objects import *
     else:
-        from .image_processing_utils import ImageProcessingMixin
+        from .image_processing_base import ImageProcessingMixin
+        from .image_processing_utils import BaseImageProcessor
         from .image_utils import ImageFeatureExtractionMixin
         from .models.beit import BeitFeatureExtractor, BeitImageProcessor
         from .models.bit import BitImageProcessor
@@ -5793,6 +5807,15 @@
         from .models.vivit import VivitImageProcessor
         from .models.yolos import YolosFeatureExtractor, YolosImageProcessor
 
+    try:
+        if not is_torchvision_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        from .utils.dummy_torchvision_objects import *
+    else:
+        from .image_processing_utils_fast import BaseImageProcessorFast
+        from .models.vit import ViTImageProcessorFast
+
     # Modeling
     try:
         if not is_torch_available():