Spaces:
Sleeping
Sleeping
File size: 10,399 Bytes
c3f3b0b |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 |
# Copyright 2023 Haotian Liu
#
# 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.
# ------------------------------------------------------------------------
# Modified from LLaVA (https://github.com/haotian-liu/LLaVA)
# Copyright 2024 Jiachen Li
# ------------------------------------------------------------------------
from PIL import Image
from io import BytesIO
import base64
import torch
import math
import ast
from transformers import StoppingCriteria
from cumo.constants import IMAGE_TOKEN_INDEX
def select_best_resolution(original_size, possible_resolutions):
"""
Selects the best resolution from a list of possible resolutions based on the original size.
Args:
original_size (tuple): The original size of the image in the format (width, height).
possible_resolutions (list): A list of possible resolutions in the format [(width1, height1), (width2, height2), ...].
Returns:
tuple: The best fit resolution in the format (width, height).
"""
original_width, original_height = original_size
best_fit = None
max_effective_resolution = 0
min_wasted_resolution = float('inf')
for width, height in possible_resolutions:
scale = min(width / original_width, height / original_height)
downscaled_width, downscaled_height = int(original_width * scale), int(original_height * scale)
effective_resolution = min(downscaled_width * downscaled_height, original_width * original_height)
wasted_resolution = (width * height) - effective_resolution
if effective_resolution > max_effective_resolution or (effective_resolution == max_effective_resolution and wasted_resolution < min_wasted_resolution):
max_effective_resolution = effective_resolution
min_wasted_resolution = wasted_resolution
best_fit = (width, height)
return best_fit
def resize_and_pad_image(image, target_resolution):
"""
Resize and pad an image to a target resolution while maintaining aspect ratio.
Args:
image (PIL.Image.Image): The input image.
target_resolution (tuple): The target resolution (width, height) of the image.
Returns:
PIL.Image.Image: The resized and padded image.
"""
original_width, original_height = image.size
target_width, target_height = target_resolution
scale_w = target_width / original_width
scale_h = target_height / original_height
if scale_w < scale_h:
new_width = target_width
new_height = min(math.ceil(original_height * scale_w), target_height)
else:
new_height = target_height
new_width = min(math.ceil(original_width * scale_h), target_width)
# Resize the image
resized_image = image.resize((new_width, new_height))
new_image = Image.new('RGB', (target_width, target_height), (0, 0, 0))
paste_x = (target_width - new_width) // 2
paste_y = (target_height - new_height) // 2
new_image.paste(resized_image, (paste_x, paste_y))
return new_image
def divide_to_patches(image, patch_size):
"""
Divides an image into patches of a specified size.
Args:
image (PIL.Image.Image): The input image.
patch_size (int): The size of each patch.
Returns:
list: A list of PIL.Image.Image objects representing the patches.
"""
patches = []
width, height = image.size
for i in range(0, height, patch_size):
for j in range(0, width, patch_size):
box = (j, i, j + patch_size, i + patch_size)
patch = image.crop(box)
patches.append(patch)
return patches
def get_anyres_image_grid_shape(image_size, grid_pinpoints, patch_size):
"""
Calculate the shape of the image patch grid after the preprocessing for images of any resolution.
Args:
image_size (tuple): The size of the input image in the format (width, height).
grid_pinpoints (str): A string representation of a list of possible resolutions.
patch_size (int): The size of each image patch.
Returns:
tuple: The shape of the image patch grid in the format (width, height).
"""
if type(grid_pinpoints) is list:
possible_resolutions = grid_pinpoints
else:
possible_resolutions = ast.literal_eval(grid_pinpoints)
width, height = select_best_resolution(image_size, possible_resolutions)
return width // patch_size, height // patch_size
def process_anyres_image(image, processor, grid_pinpoints):
"""
Process an image with variable resolutions.
Args:
image (PIL.Image.Image): The input image to be processed.
processor: The image processor object.
grid_pinpoints (str): A string representation of a list of possible resolutions.
Returns:
torch.Tensor: A tensor containing the processed image patches.
"""
if type(grid_pinpoints) is list:
possible_resolutions = grid_pinpoints
else:
possible_resolutions = ast.literal_eval(grid_pinpoints)
best_resolution = select_best_resolution(image.size, possible_resolutions)
image_padded = resize_and_pad_image(image, best_resolution)
patches = divide_to_patches(image_padded, processor.crop_size['height'])
image_original_resize = image.resize((processor.size['shortest_edge'], processor.size['shortest_edge']))
image_patches = [image_original_resize] + patches
image_patches = [processor.preprocess(image_patch, return_tensors='pt')['pixel_values'][0]
for image_patch in image_patches]
return torch.stack(image_patches, dim=0)
def load_image_from_base64(image):
return Image.open(BytesIO(base64.b64decode(image)))
def expand2square(pil_img, background_color):
width, height = pil_img.size
if width == height:
return pil_img
elif width > height:
result = Image.new(pil_img.mode, (width, width), background_color)
result.paste(pil_img, (0, (width - height) // 2))
return result
else:
result = Image.new(pil_img.mode, (height, height), background_color)
result.paste(pil_img, ((height - width) // 2, 0))
return result
def process_images(images, image_processor, model_cfg):
image_aspect_ratio = getattr(model_cfg, "image_aspect_ratio", None)
new_images = []
if image_aspect_ratio == 'pad':
for image in images:
image = expand2square(image, tuple(int(x*255) for x in image_processor.image_mean))
image = image_processor.preprocess(image, return_tensors='pt')['pixel_values'][0]
new_images.append(image)
elif image_aspect_ratio == "anyres":
for image in images:
image = process_anyres_image(image, image_processor, model_cfg.image_grid_pinpoints)
new_images.append(image)
else:
return image_processor(images, return_tensors='pt')['pixel_values']
if all(x.shape == new_images[0].shape for x in new_images):
new_images = torch.stack(new_images, dim=0)
return new_images
def tokenizer_image_token(prompt, tokenizer, image_token_index=IMAGE_TOKEN_INDEX, return_tensors=None):
prompt_chunks = [tokenizer(chunk).input_ids for chunk in prompt.split('<image>')]
def insert_separator(X, sep):
return [ele for sublist in zip(X, [sep]*len(X)) for ele in sublist][:-1]
input_ids = []
offset = 0
if len(prompt_chunks) > 0 and len(prompt_chunks[0]) > 0 and prompt_chunks[0][0] == tokenizer.bos_token_id:
offset = 1
input_ids.append(prompt_chunks[0][0])
for x in insert_separator(prompt_chunks, [image_token_index] * (offset + 1)):
input_ids.extend(x[offset:])
if return_tensors is not None:
if return_tensors == 'pt':
return torch.tensor(input_ids, dtype=torch.long)
raise ValueError(f'Unsupported tensor type: {return_tensors}')
return input_ids
def get_model_name_from_path(model_path):
model_path = model_path.strip("/")
model_paths = model_path.split("/")
if model_paths[-1].startswith('checkpoint-'):
return model_paths[-2] + "_" + model_paths[-1]
else:
return model_paths[-1]
class KeywordsStoppingCriteria(StoppingCriteria):
def __init__(self, keywords, tokenizer, input_ids):
self.keywords = keywords
self.keyword_ids = []
self.max_keyword_len = 0
for keyword in keywords:
cur_keyword_ids = tokenizer(keyword).input_ids
if len(cur_keyword_ids) > 1 and cur_keyword_ids[0] == tokenizer.bos_token_id:
cur_keyword_ids = cur_keyword_ids[1:]
if len(cur_keyword_ids) > self.max_keyword_len:
self.max_keyword_len = len(cur_keyword_ids)
self.keyword_ids.append(torch.tensor(cur_keyword_ids))
self.tokenizer = tokenizer
self.start_len = input_ids.shape[1]
def call_for_batch(self, output_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool:
offset = min(output_ids.shape[1] - self.start_len, self.max_keyword_len)
self.keyword_ids = [keyword_id.to(output_ids.device) for keyword_id in self.keyword_ids]
for keyword_id in self.keyword_ids:
truncated_output_ids = output_ids[0, -keyword_id.shape[0]:]
if torch.equal(truncated_output_ids, keyword_id):
return True
outputs = self.tokenizer.batch_decode(output_ids[:, -offset:], skip_special_tokens=True)[0]
for keyword in self.keywords:
if keyword in outputs:
return True
return False
def __call__(self, output_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool:
outputs = []
for i in range(output_ids.shape[0]):
outputs.append(self.call_for_batch(output_ids[i].unsqueeze(0), scores))
return all(outputs)
|