# Copyright 2026 the HuggingFace 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. """video processor class for KimiK2.5.""" import math import torch import torchvision.transforms.v2.functional as tvF from ...image_processing_utils import BatchFeature from ...image_utils import ( IMAGENET_STANDARD_MEAN, IMAGENET_STANDARD_STD, ChannelDimension, PILImageResampling, SizeDict, get_image_size, ) from ...processing_utils import Unpack, VideosKwargs from ...utils import TensorType from ...video_processing_utils import BaseVideoProcessor from ...video_utils import group_videos_by_shape, reorder_videos # Same resize as in image processing def navit_resize( height: int, width: int, patch_size: int, merge_kernel_size: int, max_patches: int, max_size_per_side: int, ) -> tuple[tuple[int, int], tuple[int, int]]: num_patches_h = max(1.0, height // patch_size) num_patches_w = max(1.0, width // patch_size) current_patch_count = num_patches_h * num_patches_w # Scale to satisfy total patch budget (affects both dims, hence sqrt) scale_for_total_patches = math.sqrt(max_patches / current_patch_count) # Scale to satisfy per-side patch budget scale_for_height_patches = (max_size_per_side * patch_size) / height scale_for_width_patches = (max_size_per_side * patch_size) / width # Use the most restrictive scale, never upscale scale = min(1.0, scale_for_total_patches, scale_for_height_patches, scale_for_width_patches) # Make sure the resized size doesn't go beyond predefined `max` new_height, new_width = max(1, int(height * scale)), max(1, int(width * scale)) new_height = min(new_height, max_size_per_side * patch_size) new_width = min(new_width, max_size_per_side * patch_size) # Calculate the padding to make the height and width divisible by the merge kernel size and patch size. factor = merge_kernel_size * patch_size pad_height = (factor - new_height % factor) % factor + new_height pad_width = (factor - new_width % factor) % factor + new_width return (new_height, new_width), (pad_height, pad_width) class Kimi_K25VideoProcessorInitKwargs(VideosKwargs, total=False): r""" max_patches (`int`, *optional*, defaults to `16384`): The max limit to resize the video. patch_size (`int`, *optional*, defaults to 14): The spatial patch size of the vision encoder. merge_kernel_size (`int`, *optional*, defaults to 2): The merge size of the vision encoder to llm encoder. temporal_patch_size (`int`, *optional*, defaults to 4): The temporal patch size of the vision encoder. """ max_patches: int patch_size: int merge_size: int temporal_patch_size: int class Kimi_K25VideoProcessor(BaseVideoProcessor): resample = PILImageResampling.BICUBIC size = {"max_height": 512, "max_width": 512} image_mean = IMAGENET_STANDARD_MEAN image_std = IMAGENET_STANDARD_STD do_resize = True do_rescale = True do_normalize = True do_convert_rgb = True patch_size = 14 temporal_patch_size = 4 merge_size = 2 max_patches = 4096 fps = 2 do_sample_frames = True valid_kwargs = Kimi_K25VideoProcessorInitKwargs model_input_names = ["pixel_values_videos", "video_grid_thw", "num_chunks_per_video"] def __init__(self, **kwargs: Unpack[Kimi_K25VideoProcessorInitKwargs]): super().__init__(**kwargs) def _validate_preprocess_kwargs( self, size: SizeDict | None = None, **kwargs, ) -> dict: if size is not None: if size.max_height is None or size.max_width is None or (size.max_height != size.max_width): raise ValueError( f"size must contain 'max_height' and 'max_width' keys with identical values but got {size}." ) super()._validate_preprocess_kwargs(size=size, **kwargs) def patchify( self, videos: "torch.Tensor", patch_size: int, ) -> tuple["torch.Tensor", int, int]: "Patchifies each video into flat layout of shape (`seq_len`, `patch_dim`) so we can concat dynamically shaped pixels." batch_size, num_frames, channel, resized_height, resized_width = videos.shape grid_h, grid_w = resized_height // patch_size, resized_width // patch_size patches = videos.view( batch_size, num_frames, channel, grid_h, patch_size, grid_w, patch_size, ) patches = patches.permute(0, 1, 3, 5, 2, 4, 6) flatten_patches = patches.reshape( batch_size, num_frames * grid_h * grid_w, channel, patch_size, patch_size, ) return flatten_patches, num_frames, grid_h, grid_w def _preprocess( self, videos: list["torch.Tensor"], do_resize: bool, size: SizeDict, resample: "PILImageResampling | tvF.InterpolationMode | int | None", do_rescale: bool, rescale_factor: float, do_normalize: bool, image_mean: float | list[float] | None, image_std: float | list[float] | None, patch_size: int, temporal_patch_size: int, merge_size: int, max_patches: int, return_tensors: str | TensorType | None, **kwargs, ): # Split video to chunks based on temporal patch size chunked_videos, num_chunks_per_video = [], [] for video in videos: for chunk in range(0, video.shape[0], temporal_patch_size): video_chunk = video[chunk : chunk + temporal_patch_size] chunked_videos.append(video_chunk) num_chunks_per_video.append(math.ceil(video.shape[0] / temporal_patch_size)) # Group videos by size for batched resizing and padding grouped_videos, grouped_videos_index = group_videos_by_shape(chunked_videos) resized_videos_grouped = {} for shape, stacked_videos in grouped_videos.items(): height, width = get_image_size(stacked_videos[0], channel_dim=ChannelDimension.FIRST) resized_height, resized_width = height, width if do_resize: (resized_height, resized_width), (pad_height, pad_width) = navit_resize( height, width, patch_size=patch_size, merge_kernel_size=merge_size, max_patches=max_patches, max_size_per_side=size.max_height, ) stacked_videos = self.resize( stacked_videos, size=SizeDict(height=resized_height, width=resized_width), resample=resample, ) stacked_videos = self.pad(stacked_videos, pad_size=SizeDict(height=pad_height, width=pad_width)) stacked_videos = torch.stack(stacked_videos, dim=0) resized_videos_grouped[shape] = stacked_videos resized_videos = reorder_videos(resized_videos_grouped, grouped_videos_index) # Group videos by size for further processing # Needed in case do_resize is False, or resize returns videos with different sizes grouped_videos, grouped_videos_index = group_videos_by_shape(resized_videos) processed_videos_grouped = {} processed_grids = {} for shape, stacked_videos in grouped_videos.items(): resized_height, resized_width = get_image_size(stacked_videos[0], channel_dim=ChannelDimension.FIRST) # Fused rescale and normalize stacked_videos = self.rescale_and_normalize( stacked_videos, do_rescale, rescale_factor, do_normalize, image_mean, image_std ) patches, grid_t, grid_h, grid_w = self.patchify( stacked_videos, patch_size=patch_size, ) processed_videos_grouped[shape] = patches processed_grids[shape] = [[grid_t, grid_h, grid_w]] * len(stacked_videos) processed_videos = reorder_videos(processed_videos_grouped, grouped_videos_index) processed_grids = reorder_videos(processed_grids, grouped_videos_index) pixel_values_videos = torch.cat(processed_videos, dim=0) video_grid_thw = torch.tensor(processed_grids) return BatchFeature( data={ "pixel_values_videos": pixel_values_videos, "video_grid_thw": video_grid_thw, "num_chunks_per_video": num_chunks_per_video, }, tensor_type=return_tensors, ) __all__ = ["Kimi_K25VideoProcessor"]