image_processing_qwen2_vl.py

import math
from typing import Optional, Union

import numpy as np

from transformers.image_processing_utils import BaseImageProcessor
from transformers.image_processing_base import BatchFeature
from transformers.image_transforms import (
    convert_to_rgb,
    resize,
    to_channel_dimension_format,
)
from transformers.utils.constants import OPENAI_CLIP_MEAN, OPENAI_CLIP_STD
from transformers.image_utils import (
    ChannelDimension,
    ImageInput,
    PILImageResampling,
    get_image_size,
    infer_channel_dimension_format,
    is_scaled_image,
    make_flat_list_of_images,
    to_numpy_array,
    valid_images,
    validate_preprocess_arguments,
)
from transformers.utils.generic import TensorType
from transformers.utils import logging
from transformers.video_utils import VideoInput, make_batched_videos


logger = logging.get_logger(__name__)


def smart_resize(
    height: int, width: int, factor: int = 28, min_pixels: int = 56 * 56, max_pixels: int = 14 * 14 * 4 * 1280
):
    """Rescales the image so that the following conditions are met:

    1. Both dimensions (height and width) are divisible by 'factor'.

    2. The total number of pixels is within the range ['min_pixels', 'max_pixels'].

    3. The aspect ratio of the image is maintained as closely as possible.

    """
    if max(height, width) / min(height, width) > 200:
        raise ValueError(
            f"absolute aspect ratio must be smaller than 200, got {max(height, width) / min(height, width)}"
        )
    h_bar = round(height / factor) * factor
    w_bar = round(width / factor) * factor
    if h_bar * w_bar > max_pixels:
        beta = math.sqrt((height * width) / max_pixels)
        h_bar = max(factor, math.floor(height / beta / factor) * factor)
        w_bar = max(factor, math.floor(width / beta / factor) * factor)
    elif h_bar * w_bar < min_pixels:
        beta = math.sqrt(min_pixels / (height * width))
        h_bar = math.ceil(height * beta / factor) * factor
        w_bar = math.ceil(width * beta / factor) * factor
    return h_bar, w_bar


class Qwen2VLImageProcessor(BaseImageProcessor):
    r"""
    Constructs a Qwen2-VL image processor that dynamically resizes images based on the original images.

    Args:
        do_resize (`bool`, *optional*, defaults to `True`):
            Whether to resize the image's (height, width) dimensions.
        size (`dict[str, int]`, *optional*, defaults to `{"shortest_edge": 56 * 56, "longest_edge": 28 * 28 * 1280}`):
            Size of the image after resizing. `shortest_edge` and `longest_edge` keys must be present.
        resample (`PILImageResampling`, *optional*, defaults to `Resampling.BICUBIC`):
            Resampling filter to use when resizing the image.
        do_rescale (`bool`, *optional*, defaults to `True`):
            Whether to rescale the image by the specified scale `rescale_factor`.
        rescale_factor (`int` or `float`, *optional*, defaults to `1/255`):
            Scale factor to use if rescaling the image.
        do_normalize (`bool`, *optional*, defaults to `True`):
            Whether to normalize the image.
        image_mean (`float` or `list[float]`, *optional*, defaults to `[0.48145466, 0.4578275, 0.40821073]`):
            Mean to use if normalizing the image. This is a float or list of floats for each channel in the image.
        image_std (`float` or `list[float]`, *optional*, defaults to `[0.26862954, 0.26130258, 0.27577711]`):
            Standard deviation to use if normalizing the image. This is a float or list of floats for each channel in the image.
        do_convert_rgb (`bool`, *optional*, defaults to `True`):
            Whether to convert the image to RGB.
        min_pixels (`int`, *optional*, defaults to `56 * 56`):
            The min pixels of the image to resize the image.
        max_pixels (`int`, *optional*, defaults to `28 * 28 * 1280`):
            The max pixels of the image to resize the image.
        patch_size (`int`, *optional*, defaults to 14):
            The spatial patch size of the vision encoder.
        temporal_patch_size (`int`, *optional*, defaults to 2):
            The temporal patch size of the vision encoder.
        merge_size (`int`, *optional*, defaults to 2):
            The merge size of the vision encoder to llm encoder.
    """

    model_input_names = ["pixel_values", "image_grid_thw", "pixel_values_videos", "video_grid_thw"]

    def __init__(
        self,
        do_resize: bool = True,
        size: Optional[dict[str, int]] = None,
        resample: PILImageResampling = PILImageResampling.BICUBIC,
        do_rescale: bool = True,
        rescale_factor: Union[int, 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: bool = True,
        min_pixels: Optional[int] = None,
        max_pixels: Optional[int] = None,
        patch_size: int = 14,
        temporal_patch_size: int = 2,
        merge_size: int = 2,
        focus_size: int = 2,
        **kwargs,
    ) -> None:
        super().__init__(**kwargs)
        if size is not None and ("shortest_edge" not in size or "longest_edge" not in size):
            raise ValueError("size must contain 'shortest_edge' and 'longest_edge' keys.")
        else:
            size = {"shortest_edge": 56 * 56, "longest_edge": 28 * 28 * 1280}
        # backward compatibility: override size with min_pixels and max_pixels if they are provided
        if min_pixels is not None:
            size["shortest_edge"] = min_pixels
        if max_pixels is not None:
            size["longest_edge"] = max_pixels
        self.min_pixels = size["shortest_edge"]
        self.max_pixels = size["longest_edge"]
        self.size = size

        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 if image_mean is not None else OPENAI_CLIP_MEAN
        self.image_std = image_std if image_std is not None else OPENAI_CLIP_STD

        self.patch_size = patch_size
        self.temporal_patch_size = temporal_patch_size
        self.merge_size = merge_size
        self.focus_size = focus_size
        self.do_convert_rgb = do_convert_rgb

    def _preprocess(
        self,
        images: Union[ImageInput, VideoInput],
        do_resize: Optional[bool] = None,
        size: Optional[dict[str, int]] = 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,
        patch_size: Optional[int] = None,
        temporal_patch_size: Optional[int] = None,
        merge_size: Optional[int] = None,
        focus_size: Optional[int] = None,
        do_convert_rgb: Optional[bool] = None,
        data_format: Optional[ChannelDimension] = ChannelDimension.FIRST,
        input_data_format: Optional[Union[str, ChannelDimension]] = None,
    ):
        """
        Preprocess an image or batch of images. Copy of the `preprocess` method from `CLIPImageProcessor`.

        Args:
            images (`ImageInput`):
                Image or batch of images to preprocess. Expects pixel values ranging from 0 to 255. If pixel values range from 0 to 1, set `do_rescale=False`.
            vision_info (`list[Dict]`, *optional*):
                Optional list of dictionaries containing additional information about vision inputs.
            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. `shortest_edge` and `longest_edge` keys must be present.
            resample (`PILImageResampling`, *optional*, defaults to `self.resample`):
                Resampling filter to use if resizing the image. This can be one of the `PILImageResampling` enums.
            do_rescale (`bool`, *optional*, defaults to `self.do_rescale`):
                Whether to rescale the image.
            rescale_factor (`float`, *optional*, defaults to `self.rescale_factor`):
                Scale factor to use if rescaling the image.
            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`):
                Mean to use if normalizing the image. Can be a float or a list of floats corresponding to the number of channels in the image.
            image_std (`float` or `list[float]`, *optional*, defaults to `self.image_std`):
                Standard deviation to use if normalizing the image. Can be a float or a list of floats corresponding to the number of channels in the image.
            patch_size (`int`, *optional*, defaults to `self.patch_size`):
                The spatial patch size of the vision encoder.
            temporal_patch_size (`int`, *optional*, defaults to `self.temporal_patch_size`):
                The temporal patch size of the vision encoder.
            merge_size (`int`, *optional*, defaults to `self.merge_size`):
                The merge size of the vision encoder to llm encoder.
            do_convert_rgb (`bool`, *optional*, defaults to `self.do_convert_rgb`):
                Whether to convert the image to RGB.
            data_format (`ChannelDimension`, *optional*, defaults to `ChannelDimension.FIRST`):
                The channel dimension format for the output 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.
                - Unset: Use the channel dimension format of the input image.
            input_data_format (`ChannelDimension` or `str`, *optional*):
                The channel dimension format for 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.   - `"none"` or `ChannelDimension.NONE`: image in (height, width) format.
        """
        images = make_flat_list_of_images(images) # type: ignore

        if do_convert_rgb:
            images = [convert_to_rgb(image) for image in images] # type: ignore

        # All transformations expect numpy arrays.
        images = [to_numpy_array(image) for image in images] # type: ignore

        if do_rescale and is_scaled_image(images[0]):
            logger.warning_once( # type: ignore
                "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])

        height, width = get_image_size(images[0], channel_dim=input_data_format) # type: ignore
        resized_height, resized_width = height, width
        processed_images = []
        for image in images:
            if do_resize:
                resized_height, resized_width = smart_resize(
                    height,
                    width,
                    factor=patch_size * merge_size * focus_size, # type: ignore
                    min_pixels=size["shortest_edge"], # type: ignore
                    max_pixels=size["longest_edge"], # type: ignore
                )
                image = resize(
                    image, size=(resized_height, resized_width), resample=resample, input_data_format=input_data_format
                )

            if do_rescale:
                image = self.rescale(image, scale=rescale_factor, input_data_format=input_data_format) # type: ignore

            if do_normalize:
                image = self.normalize(
                    image=image, mean=image_mean, std=image_std, input_data_format=input_data_format # type: ignore
                )

            image = to_channel_dimension_format(image, data_format, input_channel_dim=input_data_format) # type: ignore
            processed_images.append(image)

        patches = np.array(processed_images)
        if data_format == ChannelDimension.LAST:
            patches = patches.transpose(0, 3, 1, 2)
        if patches.shape[0] % temporal_patch_size != 0: # type: ignore
            repeats = np.repeat(
                patches[-1][np.newaxis], temporal_patch_size - (patches.shape[0] % temporal_patch_size), axis=0 # type: ignore
            )
            patches = np.concatenate([patches, repeats], axis=0)
        channel = patches.shape[1]
        grid_t = patches.shape[0] // temporal_patch_size # type: ignore
        grid_h, grid_w = resized_height // patch_size, resized_width // patch_size # type: ignore
        patches = patches.reshape(
            grid_t,
            temporal_patch_size, # type: ignore
            channel,
            grid_h // merge_size, # type: ignore
            merge_size, # type: ignore
            patch_size, # type: ignore
            grid_w // merge_size, # type: ignore
            merge_size, # type: ignore
            patch_size, # type: ignore
        )
        patches = patches.transpose(0, 3, 6, 4, 7, 2, 1, 5, 8)
        flatten_patches = patches.reshape(
            grid_t * grid_h * grid_w, channel * temporal_patch_size * patch_size * patch_size # type: ignore
        )

        return flatten_patches, (grid_t, grid_h, grid_w)

    def preprocess( # type: ignore
        self,
        images: ImageInput,
        videos: Optional[VideoInput] = None,
        do_resize: Optional[bool] = None,
        size: Optional[dict[str, int]] = None,
        min_pixels: Optional[int] = None,
        max_pixels: Optional[int] = 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,
        patch_size: Optional[int] = None,
        temporal_patch_size: Optional[int] = None,
        merge_size: Optional[int] = None,
        do_convert_rgb: Optional[bool] = None,
        return_tensors: Optional[Union[str, TensorType]] = None,
        data_format: Optional[ChannelDimension] = ChannelDimension.FIRST,
        input_data_format: Optional[Union[str, ChannelDimension]] = None,
    ):
        """
        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`.
            videos (`VideoInput`):
                Video to preprocess. Expects a single or batch of videos with pixel values ranging from 0 to 255. If
                passing in videos 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. Shortest edge of the image is resized to size["shortest_edge"], with
                the longest edge resized to keep the input aspect ratio.
            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`.
            min_pixels (`int`, *optional*, defaults to `self.min_pixels`):
                The min pixels of the image to resize the image.
            max_pixels (`int`, *optional*, defaults to `self.max_pixels`):
                The max pixels of the image to resize the image.
            patch_size (`int`, *optional*, defaults to `self.patch_size`):
                The spatial patch size of the vision encoder.
            temporal_patch_size (`int`, *optional*, defaults to `self.temporal_patch_size`):
                The temporal patch size of the vision encoder.
            merge_size (`int`, *optional*, defaults to `self.merge_size`):
                The merge size of the vision encoder to llm encoder.
            do_convert_rgb (`bool`, *optional*, defaults to `self.do_convert_rgb`):
                Whether to convert the image to RGB.
            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`.
            data_format (`ChannelDimension` or `str`, *optional*, defaults to `ChannelDimension.FIRST`):
                The channel dimension format for the output 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.
                - Unset: Use the channel dimension format of the input image.
            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.

        """
        min_pixels = min_pixels if min_pixels is not None else self.min_pixels
        max_pixels = max_pixels if max_pixels is not None else self.max_pixels

        if size is not None:
            if "shortest_edge" not in size or "longest_edge" not in size:
                raise ValueError("size must contain 'shortest_edge' and 'longest_edge' keys.")
            min_pixels = size["shortest_edge"]
        elif min_pixels is not None and max_pixels is not None:
            # backward compatibility: override size with min_pixels and max_pixels if they are provided
            size = {"shortest_edge": min_pixels, "longest_edge": max_pixels}
        else:
            size = {**self.size}

        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
        patch_size = patch_size if patch_size is not None else self.patch_size
        temporal_patch_size = temporal_patch_size if temporal_patch_size is not None else self.temporal_patch_size
        merge_size = merge_size if merge_size is not None else self.merge_size
        do_convert_rgb = do_convert_rgb if do_convert_rgb is not None else self.do_convert_rgb

        if images is not None:
            images = self.fetch_images(images) # type: ignore
            images = make_flat_list_of_images(images)

        if images is not None and 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(
            rescale_factor=rescale_factor,
            do_normalize=do_normalize,
            image_mean=image_mean,
            image_std=image_std,
            do_resize=do_resize,
            size=size,
            resample=resample,
        )

        data = {}
        if images is not None:
            pixel_values, vision_grid_thws = [], []
            for image in images: # type: ignore
                patches, image_grid_thw = self._preprocess(
                    image,
                    do_resize=do_resize,
                    size=size,
                    resample=resample,
                    do_rescale=do_rescale,
                    rescale_factor=rescale_factor,
                    do_normalize=do_normalize,
                    image_mean=image_mean,
                    image_std=image_std,
                    patch_size=patch_size,
                    temporal_patch_size=temporal_patch_size,
                    merge_size=merge_size,
                    data_format=data_format,
                    do_convert_rgb=do_convert_rgb,
                    input_data_format=input_data_format,
                )
                pixel_values.extend(patches)
                vision_grid_thws.append(image_grid_thw)
            pixel_values = np.array(pixel_values)
            vision_grid_thws = np.array(vision_grid_thws)
            data.update({"pixel_values": pixel_values, "image_grid_thw": vision_grid_thws})

        # kept for BC only and should be removed after v5.0
        if videos is not None:
            logger.warning(
                "`Qwen2VLImageProcessor` works only with image inputs and doesn't process videos anymore. "
                "This is a deprecated behavior and will be removed in v5.0. "
                "Your videos should be forwarded to `Qwen2VLVideoProcessor`. "
            )
            videos = make_batched_videos(videos) # type: ignore
            pixel_values_videos, vision_grid_thws_videos = [], []
            for images in videos: # type: ignore
                patches, video_grid_thw = self._preprocess(
                    images,
                    do_resize=do_resize,
                    size=size,
                    resample=resample,
                    do_rescale=do_rescale,
                    rescale_factor=rescale_factor,
                    do_normalize=do_normalize,
                    image_mean=image_mean,
                    image_std=image_std,
                    patch_size=patch_size,
                    temporal_patch_size=temporal_patch_size,
                    merge_size=merge_size,
                    data_format=data_format,
                    do_convert_rgb=do_convert_rgb,
                    input_data_format=input_data_format,
                )
                pixel_values_videos.extend(patches)
                vision_grid_thws_videos.append(video_grid_thw)
            data.update(
                {
                    "pixel_values_videos": np.array(pixel_values_videos),
                    "video_grid_thw": np.array(vision_grid_thws_videos),
                }
            )

        return BatchFeature(data=data, tensor_type=return_tensors)

    def get_number_of_image_patches(self, height: int, width: int, images_kwargs=None):
        """
        A utility that returns number of image patches for a given image size.

        Args:
            height (`int`):
                Height of the input image.
            width (`int`):
                Width of the input image.
            images_kwargs (`dict`, *optional*)
                Any kwargs to override defaults of the image processor.
        Returns:
            `int`: Number of image patches per image.
        """
        min_pixels = images_kwargs["min_pixels"] if "min_pixels" in images_kwargs else self.size["shortest_edge"] # type: ignore
        max_pixels = images_kwargs["max_pixels"] if "max_pixels" in images_kwargs else self.size["longest_edge"] # type: ignore
        patch_size = images_kwargs.get("patch_size", self.patch_size) # type: ignore
        merge_size = images_kwargs.get("merge_size", self.merge_size) # type: ignore
        focus_size = images_kwargs.get("focus_size", self.focus_size) # type: ignore

        factor = patch_size * merge_size * focus_size
        resized_height, resized_width = smart_resize(
            height, width, factor, min_pixels=min_pixels, max_pixels=max_pixels
        )
        grid_h, grid_w = resized_height // patch_size, resized_width // patch_size
        return grid_h * grid_w


__all__ = ["Qwen2VLImageProcessor"]