# 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. """Image processor class for SigLIP2.""" import math from functools import lru_cache from typing import List, Optional, Tuple, Union import numpy as np from ...image_processing_utils import BaseImageProcessor, BatchFeature from ...image_transforms import ( convert_to_rgb, resize, to_channel_dimension_format, ) from ...image_utils import ( ChannelDimension, ImageInput, PILImageResampling, infer_channel_dimension_format, is_scaled_image, make_flat_list_of_images, to_numpy_array, valid_images, validate_preprocess_arguments, ) from ...utils import TensorType, filter_out_non_signature_kwargs, is_vision_available, logging logger = logging.get_logger(__name__) if is_vision_available(): from PIL import Image @lru_cache(maxsize=256) def get_image_size_for_max_num_patches( image_height: int, image_width: int, patch_size: int, max_num_patches: int, eps: float = 1e-5 ) -> Tuple[int, int]: """ Determine image size based on max number of patches, ensure dimensions are divisible by patch size and image is at least 1 patch. Args: image_height (`int`): Original image height. image_width (`int`): Original image width. patch_size (`int`): Patch size for processing. max_num_patches (`int`): Maximum number of patches. eps (`float`): Small threshold for binary search. Returns: Tuple: (target_height, target_width) """ def get_scaled_image_size(scale: float, size: int, patch_size: int) -> int: scaled_size = size * scale scaled_size = math.ceil(scaled_size / patch_size) * patch_size # make divisible by patch_size scaled_size = max(patch_size, scaled_size) # ensure at least 1 patch return int(scaled_size) # Binary search for optimal scale scale_min, scale_max = eps / 10, 100.0 while (scale_max - scale_min) >= eps: scale = (scale_min + scale_max) / 2 target_height = get_scaled_image_size(scale, image_height, patch_size) target_width = get_scaled_image_size(scale, image_width, patch_size) num_patches = (target_height / patch_size) * (target_width / patch_size) if num_patches <= max_num_patches: scale_min = scale else: scale_max = scale scale = scale_min target_height = get_scaled_image_size(scale, image_height, patch_size) target_width = get_scaled_image_size(scale, image_width, patch_size) return target_height, target_width def convert_image_to_patches(image: np.ndarray, patch_size: int) -> np.ndarray: """ Convert 3D array image of shape (image_height, image_width, num_channels) into 2D array of patches of shape (num_patches_height * num_patches_width, patch_size * patch_size * num_channels). """ image_height, image_width, num_channels = image.shape num_patches_height = image_height // patch_size num_patches_width = image_width // patch_size patched_image = image.reshape(num_patches_height, patch_size, num_patches_width, patch_size, num_channels) patched_image = patched_image.transpose(0, 2, 1, 3, 4) patched_image = patched_image.reshape(num_patches_height * num_patches_width, -1) return patched_image def pad_along_first_dim(array: np.ndarray, target_length: int, pad_value: int = 0) -> Tuple[np.ndarray, np.ndarray]: """ Pad the array along the first dimension. """ current_length = array.shape[0] padding_length = target_length - current_length mask = np.ones((target_length,), dtype=np.int32) if padding_length > 0: paddings = [(0, padding_length)] + [(0, 0)] * (array.ndim - 1) array = np.pad(array, paddings, mode="constant", constant_values=pad_value) mask[-padding_length:] = 0 return array, mask class Siglip2ImageProcessor(BaseImageProcessor): r""" Constructs a SigLIP2 image processor. Args: do_resize (`bool`, *optional*, defaults to `True`): Whether to resize the image's dimensions to fit `max_num_patches` according to given `patch_size`. Can be overridden by `do_resize` in the `preprocess` method. resample (`PILImageResampling`, *optional*, defaults to `Resampling.BILINEAR`): Resampling filter to use if resizing the image. Can be overridden by `resample` in the `preprocess` method. do_rescale (`bool`, *optional*, defaults to `True`): Whether to rescale the image by the specified scale `rescale_factor`. Can be overridden by `do_rescale` in the `preprocess` method. rescale_factor (`int` or `float`, *optional*, defaults to `1/255`): Scale factor to use if rescaling the image. Can be overridden by `rescale_factor` in the `preprocess` method. do_normalize (`bool`, *optional*, defaults to `True`): Whether to normalize the image by the specified mean and standard deviation. Can be overridden by `do_normalize` in the `preprocess` method. image_mean (`float` or `List[float]`, *optional*, defaults to `[0.5, 0.5, 0.5]`): Mean to use if normalizing the image. This is a float or list of floats the length of the number of channels in the image. Can be overridden by the `image_mean` parameter in the `preprocess` method. image_std (`float` or `List[float]`, *optional*, defaults to `[0.5, 0.5, 0.5]`): Standard deviation to use if normalizing the image. This is a float or list of floats the length of the number of channels in the image. Can be overridden by the `image_std` parameter in the `preprocess` method. Can be overridden by the `image_std` parameter in the `preprocess` method. do_convert_rgb (`bool`, *optional*, defaults to `True`): Whether to convert the image to RGB. patch_size (`int`, *optional*, defaults to 16): The size (resolution) of each patch the image will be split to. max_num_patches (`int`, *optional*, defaults to 256): The image will be resized to have at most this number of patches, and then padded in "patch" dimension to match this number exactly. """ model_input_names = ["pixel_values", "pixel_attention_mask", "spatial_shapes"] def __init__( self, do_resize: bool = True, resample: PILImageResampling = PILImageResampling.BILINEAR, do_rescale: bool = True, rescale_factor: float = 1 / 255, do_normalize: bool = True, image_mean: Optional[Union[float, List[float]]] = None, image_std: Optional[Union[float, List[float]]] = None, do_convert_rgb: Optional[bool] = None, patch_size: int = 16, max_num_patches: int = 256, **kwargs, ): super().__init__(**kwargs) image_mean = image_mean if image_mean is not None else [0.5, 0.5, 0.5] image_std = image_std if image_std is not None else [0.5, 0.5, 0.5] self.do_resize = do_resize self.resample = resample self.do_rescale = do_rescale self.rescale_factor = rescale_factor self.do_normalize = do_normalize self.image_mean = image_mean self.image_std = image_std self.do_convert_rgb = do_convert_rgb self.patch_size = patch_size self.max_num_patches = max_num_patches @filter_out_non_signature_kwargs() def preprocess( self, images: ImageInput, do_resize: Optional[bool] = None, resample: Optional[PILImageResampling] = None, do_rescale: Optional[bool] = None, rescale_factor: Optional[float] = None, do_normalize: Optional[bool] = None, image_mean: Optional[Union[float, List[float]]] = None, image_std: Optional[Union[float, List[float]]] = None, return_tensors: Optional[Union[str, TensorType]] = None, input_data_format: Optional[Union[str, ChannelDimension]] = None, do_convert_rgb: Optional[bool] = None, patch_size: Optional[int] = None, max_num_patches: Optional[int] = None, ) -> "Image.Image": """ Preprocess an image or batch of images. Args: images (`ImageInput`): Image to preprocess. Expects a single or batch of images with pixel values ranging from 0 to 255. If passing in images with pixel values between 0 and 1, set `do_rescale=False`. do_resize (`bool`, *optional*, defaults to `self.do_resize`): Whether to resize the image. size (`Dict[str, int]`, *optional*, defaults to `self.size`): Size of the image after resizing. resample (`int`, *optional*, defaults to `self.resample`): Resampling filter to use if resizing the image. This can be one of the enum `PILImageResampling`. Only has an effect if `do_resize` is set to `True`. do_rescale (`bool`, *optional*, defaults to `self.do_rescale`): Whether to rescale the image. rescale_factor (`float`, *optional*, defaults to `self.rescale_factor`): Rescale factor to rescale the image by if `do_rescale` is set to `True`. do_normalize (`bool`, *optional*, defaults to `self.do_normalize`): Whether to normalize the image. image_mean (`float` or `List[float]`, *optional*, defaults to `self.image_mean`): Image mean to use for normalization. Only has an effect if `do_normalize` is set to `True`. image_std (`float` or `List[float]`, *optional*, defaults to `self.image_std`): Image standard deviation to use for normalization. Only has an effect if `do_normalize` is set to `True`. return_tensors (`str` or `TensorType`, *optional*): The type of tensors to return. Can be one of: - Unset: Return a list of `np.ndarray`. - `TensorType.TENSORFLOW` or `'tf'`: Return a batch of type `tf.Tensor`. - `TensorType.PYTORCH` or `'pt'`: Return a batch of type `torch.Tensor`. - `TensorType.NUMPY` or `'np'`: Return a batch of type `np.ndarray`. - `TensorType.JAX` or `'jax'`: Return a batch of type `jax.numpy.ndarray`. input_data_format (`ChannelDimension` or `str`, *optional*): The channel dimension format for the input image. If unset, the channel dimension format is inferred from the input image. Can be one of: - `"channels_first"` or `ChannelDimension.FIRST`: image in (num_channels, height, width) format. - `"channels_last"` or `ChannelDimension.LAST`: image in (height, width, num_channels) format. - `"none"` or `ChannelDimension.NONE`: image in (height, width) format. do_convert_rgb (`bool`, *optional*, defaults to `self.do_convert_rgb`): Whether to convert the image to RGB. patch_size (`int`, *optional*, defaults to `self.patch_size`): Patch size for processing, same as the patch size used in the model. max_num_patches (`int`, *optional*, defaults to `self.max_num_patches`): Maximum number of patches per image, the image will be resized to have at most this number of patches. """ do_resize = do_resize if do_resize is not None else self.do_resize resample = resample if resample is not None else self.resample do_rescale = do_rescale if do_rescale is not None else self.do_rescale rescale_factor = rescale_factor if rescale_factor is not None else self.rescale_factor do_normalize = do_normalize if do_normalize is not None else self.do_normalize image_mean = image_mean if image_mean is not None else self.image_mean image_std = image_std if image_std is not None else self.image_std do_convert_rgb = do_convert_rgb if do_convert_rgb is not None else self.do_convert_rgb patch_size = patch_size if patch_size is not None else self.patch_size max_num_patches = max_num_patches if max_num_patches is not None else self.max_num_patches # Explicitly specify data format to be channels last for image preprocessing. # Image processor does not support different output formats, because it returns patches. data_format = ChannelDimension.LAST images = make_flat_list_of_images(images) if not valid_images(images): raise ValueError( "Invalid image type. Must be of type PIL.Image.Image, numpy.ndarray, " "torch.Tensor, tf.Tensor or jax.ndarray." ) validate_preprocess_arguments( do_rescale=do_rescale, rescale_factor=rescale_factor, do_normalize=do_normalize, image_mean=image_mean, image_std=image_std, ) if do_convert_rgb: images = [convert_to_rgb(image) for image in images] # All transformations expect numpy arrays. images = [to_numpy_array(image) for image in images] if do_rescale and is_scaled_image(images[0]): logger.warning_once( "It looks like you are trying to rescale already rescaled images. If the input" " images have pixel values between 0 and 1, set `do_rescale=False` to avoid rescaling them again." ) if input_data_format is None: # We assume that all images have the same channel dimension format. input_data_format = infer_channel_dimension_format(images[0]) pixel_masks = [] pixel_values = [] spatial_shapes = [] for image in images: image = to_channel_dimension_format(image, data_format, input_channel_dim=input_data_format) if do_resize: height, width = get_image_size_for_max_num_patches( image_height=image.shape[0], image_width=image.shape[1], patch_size=patch_size, max_num_patches=max_num_patches, ) image = resize(image=image, size=(height, width), resample=resample, input_data_format=data_format) if do_rescale: image = self.rescale(image=image, scale=rescale_factor, input_data_format=data_format) if do_normalize: image = self.normalize(image=image, mean=image_mean, std=image_std, input_data_format=data_format) patches = convert_image_to_patches(image, patch_size) patches, mask = pad_along_first_dim(patches, max_num_patches) num_patches_height = image.shape[0] // patch_size num_patches_width = image.shape[1] // patch_size spatial_shapes.append((num_patches_height, num_patches_width)) pixel_values.append(patches) pixel_masks.append(mask) batch_feature = BatchFeature( data={ "pixel_values": pixel_values, "pixel_attention_mask": pixel_masks, "spatial_shapes": spatial_shapes, }, tensor_type=return_tensors, ) return batch_feature __all__ = ["Siglip2ImageProcessor"]