add initial design for uniform processors + align model

src/transformers/models/align/processing_align.py

@@ -16,8 +16,66 @@
 Image/Text processor class for ALIGN
 """
 
-from ...processing_utils import ProcessorMixin
-from ...tokenization_utils_base import BatchEncoding
+from typing import List, Union
+
+from ...image_utils import ImageInput
+from ...processing_utils import (
+    CommonKwargs,
+    ImagesKwargs,
+    ProcessingKwargs,
+    ProcessorMixin,
+    TextKwargs,
+)
+from ...tokenization_utils_base import BatchEncoding, PreTokenizedInput, TextInput
+from ...utils import is_torch_available, is_vision_available
+
+
+# TODO (@molbap) This is a bother, forward references from TypedDict are resolved and need this to work
+if is_vision_available():
+    import PIL  # noqa: F401
+if is_torch_available():
+    import torch  # noqa: F401
+
+
+class AlignProcessorKwargs(ProcessingKwargs, total=False):
+    """
+    Inherits from `ProcessingKwargs` to provide:
+        1) Additional keys that this model requires to process inputs.
+        2) Default values for extra keys.
+    New keys have to be defined as follows to ensure type hinting is done correctly.
+
+    ```python
+    common_kwargs: CommonKwargs = {
+            **CommonKwargs.__annotations__,
+        }
+        text_kwargs: TextKwargs = {
+            **TextKwargs.__annotations__,
+            "a_new_text_boolean_key": Optional[bool],
+        }
+        images_kwargs: ImagesKwargs = {
+            **ImagesKwargs.__annotations__,
+            "a_new_image_processing_key": Optional[int]
+        }
+    ```
+
+    """
+
+    common_kwargs: CommonKwargs = {
+        **CommonKwargs.__annotations__,
+    }
+    text_kwargs: TextKwargs = {
+        **TextKwargs.__annotations__,
+    }
+    images_kwargs: ImagesKwargs = {
+        **ImagesKwargs.__annotations__,
+    }
+
+    _defaults = {
+        "text_kwargs": {
+            "padding": "max_length",
+            "max_length": 64,
+        },
+    }
 
 
 class AlignProcessor(ProcessorMixin):
@@ -26,12 +84,39 @@ class AlignProcessor(ProcessorMixin):
     [`BertTokenizer`]/[`BertTokenizerFast`] into a single processor that interits both the image processor and
     tokenizer functionalities. See the [`~AlignProcessor.__call__`] and [`~OwlViTProcessor.decode`] for more
     information.
+    The preferred way of passing kwargs is as a dictionary per modality, see usage example below.
+        ```python
+        from transformers import AlignProcessor
+        from PIL import Image
+        model_id = "kakaobrain/align-base"
+        processor = AlignProcessor.from_pretrained(model_id)
+
+        # Define the kwargs for each modality
+        common_kwargs = {"return_tensors": "pt"}
+        images_kwargs = {"crop_size": {"height": 224, "width": 224}}
+        text_kwargs = {"padding": "do_not_pad"}
+
+        # Combine them into a single dictionary
+
+        all_kwargs = {
+        "images_kwargs": images_kwargs,
+        "text_kwargs": text_kwargs,
+        "common_kwargs": common_kwargs
+        }
+
+        processor(images=your_pil_image, text=["What is that?"], **all_kwargs)
+
+        # passing directly any number of kwargs is also supported, but not recommended
+
+        processor(images=your_pil_image, text=["What is that?"], padding="do_not_pad)
+        ```
 
     Args:
         image_processor ([`EfficientNetImageProcessor`]):
             The image processor is a required input.
         tokenizer ([`BertTokenizer`, `BertTokenizerFast`]):
             The tokenizer is a required input.
+
     """
 
     attributes = ["image_processor", "tokenizer"]
@@ -41,11 +126,21 @@ class AlignProcessor(ProcessorMixin):
     def __init__(self, image_processor, tokenizer):
         super().__init__(image_processor, tokenizer)
 
-    def __call__(self, text=None, images=None, padding="max_length", max_length=64, return_tensors=None, **kwargs):
+    def __call__(
+        self,
+        text: Union[TextInput, PreTokenizedInput, List[TextInput], List[PreTokenizedInput]] = None,
+        images: ImageInput = None,
+        audio=None,
+        videos=None,
+        text_kwargs: AlignProcessorKwargs.text_kwargs = None,
+        images_kwargs: AlignProcessorKwargs.images_kwargs = None,
+        common_kwargs: AlignProcessorKwargs.common_kwargs = None,
+        **kwargs: AlignProcessorKwargs,
+    ) -> BatchEncoding:
         """
         Main method to prepare text(s) and image(s) to be fed as input to the model. This method forwards the `text`
-        and `kwargs` arguments to BertTokenizerFast's [`~BertTokenizerFast.__call__`] if `text` is not `None` to encode
-        the text. To prepare the image(s), this method forwards the `images` and `kwargs` arguments to
+        arguments to BertTokenizerFast's [`~BertTokenizerFast.__call__`] if `text` is not `None` to encode
+        the text. To prepare the image(s), this method forwards the `images` arguments to
         EfficientNetImageProcessor's [`~EfficientNetImageProcessor.__call__`] if `images` is not `None`. Please refer
         to the doctsring of the above two methods for more information.
 
@@ -57,20 +152,12 @@ def __call__(self, text=None, images=None, padding="max_length", max_length=64,
             images (`PIL.Image.Image`, `np.ndarray`, `torch.Tensor`, `List[PIL.Image.Image]`, `List[np.ndarray]`, `List[torch.Tensor]`):
                 The image or batch of images to be prepared. Each image can be a PIL image, NumPy array or PyTorch
                 tensor. Both channels-first and channels-last formats are supported.
-            padding (`bool`, `str` or [`~utils.PaddingStrategy`], *optional*, defaults to `max_length`):
-                Activates and controls padding for tokenization of input text. Choose between [`True` or `'longest'`,
-                `'max_length'`, `False` or `'do_not_pad'`]
-            max_length (`int`, *optional*, defaults to `max_length`):
-                Maximum padding value to use to pad the input text during tokenization.
-
             return_tensors (`str` or [`~utils.TensorType`], *optional*):
                 If set, will return tensors of a particular framework. Acceptable values are:
-
-                - `'tf'`: Return TensorFlow `tf.constant` objects.
-                - `'pt'`: Return PyTorch `torch.Tensor` objects.
-                - `'np'`: Return NumPy `np.ndarray` objects.
-                - `'jax'`: Return JAX `jnp.ndarray` objects.
-
+                    - `'tf'`: Return TensorFlow `tf.constant` objects.
+                    - `'pt'`: Return PyTorch `torch.Tensor` objects.
+                    - `'np'`: Return NumPy `np.ndarray` objects.
+                    - `'jax'`: Return JAX `jnp.ndarray` objects.
         Returns:
             [`BatchEncoding`]: A [`BatchEncoding`] with the following fields:
 
@@ -81,15 +168,53 @@ def __call__(self, text=None, images=None, padding="max_length", max_length=64,
             - **pixel_values** -- Pixel values to be fed to a model. Returned when `images` is not `None`.
         """
         if text is None and images is None:
-            raise ValueError("You have to specify either text or images. Both cannot be none.")
-
+            raise ValueError("You must specify either text or images.")
+        # set kwargs as empty dicts to avoid default mutable
+        if text_kwargs is None:
+            text_kwargs = {}
+        if images_kwargs is None:
+            images_kwargs = {}
+        if common_kwargs is None:
+            common_kwargs = {}
+        # Init with default values if they exist
+        default_text_kwargs = AlignProcessorKwargs._defaults.get("text_kwargs", {}).copy()
+
+        # then override with tokenizer-level arguments passed
+        default_text_kwargs.update(
+            {k: v for k, v in self.tokenizer.init_kwargs.items() if k in AlignProcessorKwargs.text_kwargs}
+        )
+        # then get passed per-modality dictionaries if they exist
+        text_kwargs = {**default_text_kwargs, **text_kwargs, **kwargs.pop("text_kwargs", {})}
+        images_kwargs.update(kwargs.pop("images_kwargs", {}))
+        common_kwargs.update(kwargs.pop("common_kwargs", {}))
+
+        # then merge kwargs by name
+        for text_key in AlignProcessorKwargs.text_kwargs.keys():
+            text_kwarg_value = kwargs.pop(text_key, None)
+            if text_kwarg_value is not None:
+                text_kwargs[text_key] = text_kwarg_value
+
+        for images_key in AlignProcessorKwargs.images_kwargs.keys():
+            images_kwarg_value = kwargs.pop(images_key, None)
+            if images_kwarg_value is not None:
+                images_kwargs[images_key] = images_kwarg_value
+        # if something remains in kwargs, it belongs to common
+        common_kwargs.update(kwargs)
+
+        # all modality-specific kwargs are updated with common kwargs
+        text_kwargs.update(common_kwargs)
+        images_kwargs.update(common_kwargs)
+
+        # then, we can pass correct kwargs to each processor
         if text is not None:
-            encoding = self.tokenizer(
-                text, padding=padding, max_length=max_length, return_tensors=return_tensors, **kwargs
-            )
+            encoding = self.tokenizer(text, **text_kwargs)
 
         if images is not None:
-            image_features = self.image_processor(images, return_tensors=return_tensors, **kwargs)
+            image_features = self.image_processor(images, **images_kwargs)
+
+        # BC for explicit return_tensors
+        if "return_tensors" in common_kwargs:
+            return_tensors = common_kwargs.pop("return_tensors", None)
 
         if text is not None and images is not None:
             encoding["pixel_values"] = image_features.pixel_values