# coding=utf-8 # Copyright 2023 The HuggingFace Inc. team. # # 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. """ Processor class for BLIP-2. """ from typing import Optional, Union from ...image_processing_utils import BatchFeature from ...image_utils import ImageInput from ...processing_utils import ProcessingKwargs, ProcessorMixin, Unpack from ...tokenization_utils_base import AddedToken, BatchEncoding, PreTokenizedInput, TextInput from ...utils import logging logger = logging.get_logger(__name__) class Blip2ProcessorKwargs(ProcessingKwargs, total=False): _defaults = { "text_kwargs": { "add_special_tokens": True, "padding": False, "stride": 0, "return_overflowing_tokens": False, "return_special_tokens_mask": False, "return_offsets_mapping": False, "return_token_type_ids": False, "return_length": False, "verbose": True, }, "images_kwargs": {}, } class Blip2Processor(ProcessorMixin): r""" Constructs a BLIP-2 processor which wraps a BLIP image processor and an OPT/T5 tokenizer into a single processor. [`BlipProcessor`] offers all the functionalities of [`BlipImageProcessor`] and [`AutoTokenizer`]. See the docstring of [`~BlipProcessor.__call__`] and [`~BlipProcessor.decode`] for more information. Args: image_processor (`BlipImageProcessor`): An instance of [`BlipImageProcessor`]. The image processor is a required input. tokenizer (`AutoTokenizer`): An instance of ['PreTrainedTokenizer`]. The tokenizer is a required input. num_query_tokens (`int`, *optional*): Number of tokens used by the Qformer as queries, should be same as in model's config. """ attributes = ["image_processor", "tokenizer"] image_processor_class = ("BlipImageProcessor", "BlipImageProcessorFast") tokenizer_class = "AutoTokenizer" def __init__(self, image_processor, tokenizer, num_query_tokens=None, **kwargs): tokenizer.return_token_type_ids = False self.current_processor = image_processor if not hasattr(tokenizer, "image_token"): self.image_token = AddedToken("", normalized=False, special=True) tokenizer.add_tokens([self.image_token], special_tokens=True) else: self.image_token = tokenizer.image_token self.num_query_tokens = num_query_tokens super().__init__(image_processor, tokenizer) def __call__( self, images: Optional[ImageInput] = None, text: Optional[Union[str, list[str], TextInput, PreTokenizedInput]] = None, audio=None, videos=None, **kwargs: Unpack[Blip2ProcessorKwargs], ) -> BatchEncoding: """ This method uses [`BlipImageProcessor.__call__`] method to prepare image(s) for the model, and [`BertTokenizerFast.__call__`] to prepare text for the model. Please refer to the docstring of the above two methods for more information. Args: images (`ImageInput`): 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. text (`TextInput`, `PreTokenizedInput`, `list[TextInput]`, `list[PreTokenizedInput]`): The sequence or batch of sequences to be encoded. Each sequence can be a string or a list of strings (pretokenized string). If the sequences are provided as list of strings (pretokenized), you must set `is_split_into_words=True` (to lift the ambiguity with a batch of sequences). 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. """ if images is None and text is None: raise ValueError("You have to specify either images or text.") output_kwargs = self._merge_kwargs( Blip2ProcessorKwargs, tokenizer_init_kwargs=self.tokenizer.init_kwargs, **kwargs, ) # BC for explicit return_tensors return_tensors = output_kwargs["text_kwargs"].pop("return_tensors", None) max_length = output_kwargs["text_kwargs"].pop("max_length", None) if max_length is not None: output_kwargs["text_kwargs"]["max_length"] = max_length - self.num_query_tokens encoding = BatchFeature(tensor_type=return_tensors) if text is not None: if isinstance(text, str): text = [text] elif not isinstance(text, list) and not isinstance(text[0], str): raise ValueError("Invalid input text. Please provide a string, or a list of strings") # We need this hacky manipulation because BLIP expects image tokens to be at the beginning even before BOS token text_encoding = self.tokenizer(text, **output_kwargs["text_kwargs"]) if images is not None and self.num_query_tokens is not None: # Image tokens should not be padded/truncated or prepended with special BOS token image_tokens = self.image_token.content * self.num_query_tokens output_kwargs["text_kwargs"]["add_special_tokens"] = False output_kwargs["text_kwargs"]["padding"] = False output_kwargs["text_kwargs"]["truncation"] = False image_text_encoding = self.tokenizer(image_tokens, **output_kwargs["text_kwargs"]) for k in text_encoding: text_encoding[k] = [image_text_encoding[k] + sample for sample in text_encoding[k]] encoding.update(text_encoding) # Now add pixel_values encoding. If we also have text_encoding, update image encoding and return it. # else, return the text encoding. if images is not None: image_encoding = self.image_processor(images, **output_kwargs["images_kwargs"]) encoding.update(image_encoding) # Cast to desired return tensors type encoding = BatchFeature(encoding, tensor_type=return_tensors) return encoding __all__ = ["Blip2Processor"]