Upload image_preprocessing_molmo.py with huggingface_hub
Browse files- image_preprocessing_molmo.py +548 -0
image_preprocessing_molmo.py
ADDED
|
@@ -0,0 +1,548 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""Image processor class for Molmo"""
|
| 2 |
+
from typing import List, Optional, Union, Mapping
|
| 3 |
+
|
| 4 |
+
import numpy as np
|
| 5 |
+
import einops
|
| 6 |
+
import torch
|
| 7 |
+
import torchvision.transforms
|
| 8 |
+
from torchvision.transforms import InterpolationMode
|
| 9 |
+
from torchvision.transforms.functional import convert_image_dtype
|
| 10 |
+
|
| 11 |
+
from transformers.image_utils import (
|
| 12 |
+
OPENAI_CLIP_MEAN,
|
| 13 |
+
OPENAI_CLIP_STD,
|
| 14 |
+
ImageInput,
|
| 15 |
+
is_valid_image,
|
| 16 |
+
)
|
| 17 |
+
from transformers.processing_utils import ImagesKwargs
|
| 18 |
+
from transformers.image_processing_utils import BaseImageProcessor
|
| 19 |
+
from transformers.utils import logging
|
| 20 |
+
|
| 21 |
+
|
| 22 |
+
logger = logging.get_logger(__name__)
|
| 23 |
+
|
| 24 |
+
|
| 25 |
+
def pad_to_bounding_box(
|
| 26 |
+
image, offset_height, offset_width, target_height,
|
| 27 |
+
target_width, value=0
|
| 28 |
+
):
|
| 29 |
+
height, width = image.shape[:2]
|
| 30 |
+
after_padding_width = target_width - offset_width - width
|
| 31 |
+
after_padding_height = target_height - offset_height - height
|
| 32 |
+
return np.pad(image, [
|
| 33 |
+
[offset_height, after_padding_height],
|
| 34 |
+
[offset_width, after_padding_width],
|
| 35 |
+
[0, 0]
|
| 36 |
+
], constant_values=value)
|
| 37 |
+
|
| 38 |
+
|
| 39 |
+
def normalize_image(image, offset, scale):
|
| 40 |
+
image -= np.array(offset, dtype=np.float32)[None, None, :]
|
| 41 |
+
image /= np.array(scale, dtype=np.float32)[None, None, :]
|
| 42 |
+
return image
|
| 43 |
+
|
| 44 |
+
|
| 45 |
+
def resize_and_pad(
|
| 46 |
+
image,
|
| 47 |
+
desired_output_size,
|
| 48 |
+
resize_method="torch-bilinear",
|
| 49 |
+
pad_value=0,
|
| 50 |
+
normalize=True,
|
| 51 |
+
image_mean=OPENAI_CLIP_MEAN,
|
| 52 |
+
image_std=OPENAI_CLIP_STD,
|
| 53 |
+
):
|
| 54 |
+
desired_height, desired_width = desired_output_size
|
| 55 |
+
height, width = image.shape[:2]
|
| 56 |
+
|
| 57 |
+
# Cast into float32 since the training code did this in float32 and it (very rarely) effects
|
| 58 |
+
# the results after rounding.
|
| 59 |
+
image_scale_y = np.array(desired_height, np.float32) / np.array(height, np.float32)
|
| 60 |
+
image_scale_x = np.array(desired_width, np.float32) / np.array(width, np.float32)
|
| 61 |
+
image_scale = min(image_scale_x, image_scale_y)
|
| 62 |
+
scaled_height = int(np.array(height, np.float32) * image_scale)
|
| 63 |
+
scaled_width = int(np.array(width, np.float32) * image_scale)
|
| 64 |
+
|
| 65 |
+
if resize_method == "tensorflow":
|
| 66 |
+
# This how the original training code did resizing, it can produce slightly different
|
| 67 |
+
# results then using torch resize so we keep it just in case
|
| 68 |
+
import tensorflow as tf
|
| 69 |
+
image = tf.image.convert_image_dtype(tf.constant(image), dtype=tf.float32)
|
| 70 |
+
image = tf.image.resize(
|
| 71 |
+
image,
|
| 72 |
+
[scaled_height, scaled_width],
|
| 73 |
+
method=tf.image.ResizeMethod.BILINEAR,
|
| 74 |
+
antialias=True,
|
| 75 |
+
)
|
| 76 |
+
image = tf.clip_by_value(image, 0.0, 1.0)
|
| 77 |
+
image = image.numpy()
|
| 78 |
+
elif resize_method == "torch-bilinear":
|
| 79 |
+
image = torch.permute(torch.from_numpy(image), [2, 0, 1])
|
| 80 |
+
image = convert_image_dtype(image) # resize in float32 to match the training code
|
| 81 |
+
image = torchvision.transforms.Resize(
|
| 82 |
+
[scaled_height, scaled_width], InterpolationMode.BILINEAR, antialias=True
|
| 83 |
+
)(image)
|
| 84 |
+
image = torch.clip(image, 0.0, 1.0)
|
| 85 |
+
image = torch.permute(image, [1, 2, 0]).numpy()
|
| 86 |
+
else:
|
| 87 |
+
raise NotImplementedError(resize_method)
|
| 88 |
+
|
| 89 |
+
top_pad = (desired_height - scaled_height) // 2
|
| 90 |
+
left_pad = (desired_width - scaled_width) // 2
|
| 91 |
+
padding = [
|
| 92 |
+
[top_pad, desired_height - scaled_height - top_pad],
|
| 93 |
+
[left_pad, desired_width - scaled_width - left_pad],
|
| 94 |
+
[0, 0]
|
| 95 |
+
]
|
| 96 |
+
image_mask = np.pad(np.ones_like(image[:, :, 0], dtype=bool), padding[:2])
|
| 97 |
+
image = np.pad(image, padding, constant_values=pad_value)
|
| 98 |
+
if normalize:
|
| 99 |
+
image = normalize_image(image, offset=image_mean, scale=image_std)
|
| 100 |
+
return image, image_mask
|
| 101 |
+
|
| 102 |
+
|
| 103 |
+
def select_tiling(h, w, patch_size, max_num_patches):
|
| 104 |
+
"""Decide how best to divide in image of size [w, h] in up to max_num_patches of size patch_size"""
|
| 105 |
+
original_size = np.stack([h, w]) # [1, 2]
|
| 106 |
+
original_res = h * w
|
| 107 |
+
tilings = []
|
| 108 |
+
for i in range(1, max_num_patches+1):
|
| 109 |
+
for j in range(1, max_num_patches+1):
|
| 110 |
+
if i*j <= max_num_patches:
|
| 111 |
+
tilings.append((i, j))
|
| 112 |
+
# sort so argmin and argmax favour smaller tilings in the event of a tie
|
| 113 |
+
tilings.sort(key=lambda x: (x[0]*x[1], x[0]))
|
| 114 |
+
candidate_tilings = np.array(tilings, dtype=np.int32) # [n_resolutions, 2]
|
| 115 |
+
candidate_resolutions = candidate_tilings * patch_size # [n_resolutions, 2]
|
| 116 |
+
|
| 117 |
+
# How much we would need to scale the image to fit exactly in each tiling
|
| 118 |
+
original_size = np.stack([h, w], dtype=np.float32) # [1, 2]
|
| 119 |
+
required_scale_d = candidate_resolutions.astype(np.float32) / original_size
|
| 120 |
+
required_scale = np.min(required_scale_d, axis=-1, keepdims=True) # [n_resolutions, 1]
|
| 121 |
+
if np.all(required_scale < 1):
|
| 122 |
+
# We are forced to downscale, so try to minimize the amount of downscaling
|
| 123 |
+
ix = np.argmax(required_scale)
|
| 124 |
+
else:
|
| 125 |
+
# Pick the resolution that required the least upscaling so that it most closely fits the image
|
| 126 |
+
required_scale = np.where(required_scale < 1.0, 10e9, required_scale)
|
| 127 |
+
ix = np.argmin(required_scale)
|
| 128 |
+
return candidate_tilings[ix]
|
| 129 |
+
|
| 130 |
+
|
| 131 |
+
class MolmoImagesKwargs(ImagesKwargs, total=False):
|
| 132 |
+
max_crops: Optional[int]
|
| 133 |
+
overlap_margins: Optional[List[int]]
|
| 134 |
+
base_image_input_size: Optional[List[int]]
|
| 135 |
+
image_token_length_w: Optional[int]
|
| 136 |
+
image_token_length_h: Optional[int]
|
| 137 |
+
image_patch_size: Optional[int]
|
| 138 |
+
image_padding_mask: Optional[bool]
|
| 139 |
+
|
| 140 |
+
|
| 141 |
+
class MolmoImageProcessor(BaseImageProcessor):
|
| 142 |
+
"""Preprocess images and multi-model inputs"""
|
| 143 |
+
|
| 144 |
+
def __init__(
|
| 145 |
+
self,
|
| 146 |
+
max_crops: int = 12,
|
| 147 |
+
overlap_margins: List[int] = (4, 4),
|
| 148 |
+
base_image_input_size: List[int] = (336, 336),
|
| 149 |
+
image_token_length_w: int = 12,
|
| 150 |
+
image_token_length_h: int = 12,
|
| 151 |
+
image_patch_size: int = 14,
|
| 152 |
+
image_padding_mask: bool = True,
|
| 153 |
+
do_normalize: bool = True,
|
| 154 |
+
image_mean: Optional[Union[float, List[float]]] = None,
|
| 155 |
+
image_std: Optional[Union[float, List[float]]] = None,
|
| 156 |
+
**kwargs,
|
| 157 |
+
):
|
| 158 |
+
super().__init__(**kwargs)
|
| 159 |
+
self.max_crops = max_crops
|
| 160 |
+
self.overlap_margins = overlap_margins
|
| 161 |
+
self.base_image_input_size = base_image_input_size
|
| 162 |
+
self.image_token_length_w = image_token_length_w
|
| 163 |
+
self.image_token_length_h = image_token_length_h
|
| 164 |
+
self.image_patch_size = image_patch_size
|
| 165 |
+
self.image_padding_mask = image_padding_mask
|
| 166 |
+
self.do_normalize = do_normalize
|
| 167 |
+
self.image_mean = image_mean if image_mean is not None else OPENAI_CLIP_MEAN
|
| 168 |
+
self.image_std = image_std if image_std is not None else OPENAI_CLIP_STD
|
| 169 |
+
|
| 170 |
+
def image_to_patches_and_tokens(
|
| 171 |
+
self,
|
| 172 |
+
image: ImageInput,
|
| 173 |
+
image_patch_token_id: int,
|
| 174 |
+
image_col_token_id: int,
|
| 175 |
+
image_start_token_id: int,
|
| 176 |
+
image_end_token_id: int,
|
| 177 |
+
max_crops: Optional[int] = None,
|
| 178 |
+
overlap_margins: Optional[List[int]] = None,
|
| 179 |
+
base_image_input_size: Optional[Union[int, List[int]]] = None,
|
| 180 |
+
image_token_length_w: Optional[int] = None,
|
| 181 |
+
image_token_length_h: Optional[int] = None,
|
| 182 |
+
image_patch_size: Optional[int] = None,
|
| 183 |
+
):
|
| 184 |
+
if isinstance(base_image_input_size, int):
|
| 185 |
+
base_image_input_size = (base_image_input_size, base_image_input_size)
|
| 186 |
+
|
| 187 |
+
base_image_input_d = image_patch_size
|
| 188 |
+
tokens_per_image = image_token_length_w * image_token_length_h
|
| 189 |
+
image_base_patch_w = base_image_input_size[1] // base_image_input_d
|
| 190 |
+
image_base_patch_h = base_image_input_size[0] // base_image_input_d
|
| 191 |
+
|
| 192 |
+
original_image_h, original_image_w = image.shape[:2]
|
| 193 |
+
crop_size = base_image_input_size[0]
|
| 194 |
+
|
| 195 |
+
# Discard this many patches from the (left/top, right/bottom) of crops
|
| 196 |
+
left_margin, right_margin = overlap_margins
|
| 197 |
+
# left_margin, right_margin = 2, 2
|
| 198 |
+
assert left_margin % 2 == 0 # Required for compatibility with 2x2 pooling
|
| 199 |
+
total_margin_pixels = base_image_input_d*(right_margin + left_margin) # pixels removed per dim
|
| 200 |
+
crop_patches = base_image_input_size[0] // base_image_input_d # patches per crop dim
|
| 201 |
+
crop_window_patches = crop_patches - (right_margin + left_margin) # usable patches
|
| 202 |
+
crop_window_size = crop_window_patches * base_image_input_d
|
| 203 |
+
tiling = select_tiling(
|
| 204 |
+
original_image_h - total_margin_pixels,
|
| 205 |
+
original_image_w - total_margin_pixels,
|
| 206 |
+
crop_window_size,
|
| 207 |
+
max_crops
|
| 208 |
+
)
|
| 209 |
+
src, img_mask = resize_and_pad(
|
| 210 |
+
image,
|
| 211 |
+
[tiling[0]*crop_window_size+total_margin_pixels, tiling[1]*crop_window_size+total_margin_pixels]
|
| 212 |
+
)
|
| 213 |
+
|
| 214 |
+
# Now we have to split the image into crops, while keeping track of how each patch in the
|
| 215 |
+
# each crop should be ordered in the global image, this require a lot of tricky booking
|
| 216 |
+
n_crops = tiling[0] * tiling[1]
|
| 217 |
+
patches_arr = []
|
| 218 |
+
mask_arr = []
|
| 219 |
+
patch_ordering_arr = []
|
| 220 |
+
|
| 221 |
+
# We assume 2x2 pooling, but can allow padding the right/bottom with extra
|
| 222 |
+
# patches if the number of patches per side is not even
|
| 223 |
+
assert (crop_patches+1)//2 == image_token_length_h
|
| 224 |
+
assert (crop_patches+1)//2 == image_token_length_w
|
| 225 |
+
on = 0
|
| 226 |
+
on_patch = 0
|
| 227 |
+
for i in range(tiling[0]):
|
| 228 |
+
y0 = i*crop_window_size
|
| 229 |
+
if i == 0:
|
| 230 |
+
crop_y0 = 0
|
| 231 |
+
else:
|
| 232 |
+
crop_y0 = left_margin // 2
|
| 233 |
+
|
| 234 |
+
crop_h = image_base_patch_h - (right_margin + left_margin)
|
| 235 |
+
if i == 0:
|
| 236 |
+
crop_h += left_margin
|
| 237 |
+
if i == (tiling[0]-1):
|
| 238 |
+
crop_h += right_margin
|
| 239 |
+
for j in range(tiling[1]):
|
| 240 |
+
x0 = j*crop_window_size
|
| 241 |
+
if j == 0:
|
| 242 |
+
crop_x0 = 0
|
| 243 |
+
else:
|
| 244 |
+
crop_x0 = left_margin // 2
|
| 245 |
+
|
| 246 |
+
crop_w = image_base_patch_w - (right_margin + left_margin)
|
| 247 |
+
if j == 0:
|
| 248 |
+
crop_w += left_margin
|
| 249 |
+
if j == (tiling[1]-1):
|
| 250 |
+
crop_w += right_margin
|
| 251 |
+
|
| 252 |
+
pooled_w = (crop_w + 1) // 2
|
| 253 |
+
pooled_h = (crop_h + 1) // 2
|
| 254 |
+
patch_ordering_arr.append(
|
| 255 |
+
pad_to_bounding_box(
|
| 256 |
+
np.reshape(np.arange(on, on+pooled_h*pooled_w, dtype=np.int32), (pooled_h, pooled_w, 1)),
|
| 257 |
+
crop_y0, crop_x0, image_token_length_h, image_token_length_w, value=-1
|
| 258 |
+
)[:, :, 0]
|
| 259 |
+
)
|
| 260 |
+
patches_arr.append(src[y0:y0+crop_size, x0:x0+crop_size])
|
| 261 |
+
mask_arr.append(img_mask[y0:y0+crop_size, x0:x0+crop_size])
|
| 262 |
+
|
| 263 |
+
on += pooled_h*pooled_w
|
| 264 |
+
on_patch += 1
|
| 265 |
+
patches = np.stack(patches_arr)
|
| 266 |
+
patch_ordering = np.stack(patch_ordering_arr)
|
| 267 |
+
img_mask = np.stack(mask_arr)
|
| 268 |
+
|
| 269 |
+
# Switch to [n_crops, n_patches, pixels_per_patch] format
|
| 270 |
+
image_layout_impatch_w, image_layout_impatch_h = tiling[0], tiling[1]
|
| 271 |
+
patches = einops.rearrange(
|
| 272 |
+
patches, 'p (h dh) (w dw) c -> p (h w) (dh dw c)',
|
| 273 |
+
dh=base_image_input_d,
|
| 274 |
+
dw=base_image_input_d,
|
| 275 |
+
h=image_base_patch_h,
|
| 276 |
+
w=image_base_patch_w
|
| 277 |
+
)
|
| 278 |
+
img_mask = einops.rearrange(
|
| 279 |
+
img_mask, 'p (h dh) (w dw) -> p (h w) (dh dw)',
|
| 280 |
+
dh=base_image_input_d,
|
| 281 |
+
dw=base_image_input_d,
|
| 282 |
+
h=image_base_patch_h,
|
| 283 |
+
w=image_base_patch_w
|
| 284 |
+
)
|
| 285 |
+
|
| 286 |
+
img_mask = img_mask.astype(np.float32).mean(axis=-1)
|
| 287 |
+
patch_ordering = np.reshape(patch_ordering, [-1])
|
| 288 |
+
valid = patch_ordering >= 0
|
| 289 |
+
|
| 290 |
+
# Transpose order, to get left-to-right order instead of crop-by-crop order
|
| 291 |
+
patch_ordering_rh = np.reshape(
|
| 292 |
+
patch_ordering,
|
| 293 |
+
[tiling[0], tiling[1], image_token_length_h, image_token_length_w]
|
| 294 |
+
)
|
| 295 |
+
patch_ordering_rh = np.transpose(patch_ordering_rh, [0, 2, 1, 3])
|
| 296 |
+
patch_ordering_rh = np.reshape(patch_ordering_rh, [-1])
|
| 297 |
+
|
| 298 |
+
# The transpose will screw up which patches are masked, project the
|
| 299 |
+
# new order into sparse structure of `patch_ordering` to fix this
|
| 300 |
+
patch_ordering[valid] = patch_ordering_rh[patch_ordering_rh >= 0]
|
| 301 |
+
|
| 302 |
+
# Now build the output tokens
|
| 303 |
+
h = tiling[0] * crop_window_patches + (right_margin+left_margin)
|
| 304 |
+
w = tiling[1] * crop_window_patches + (right_margin+left_margin)
|
| 305 |
+
per_row = np.full(
|
| 306 |
+
((w+1)//2,),
|
| 307 |
+
image_patch_token_id,
|
| 308 |
+
)
|
| 309 |
+
per_row = np.concatenate([per_row, [image_col_token_id]], 0)
|
| 310 |
+
|
| 311 |
+
joint = np.tile(per_row, [(h+1)//2])
|
| 312 |
+
joint = [
|
| 313 |
+
[image_start_token_id],
|
| 314 |
+
joint,
|
| 315 |
+
[image_end_token_id]
|
| 316 |
+
]
|
| 317 |
+
|
| 318 |
+
# Finally do the same for the global image
|
| 319 |
+
resized, _ = resize_and_pad(image, base_image_input_size)
|
| 320 |
+
resized = einops.rearrange(
|
| 321 |
+
resized, '(h dh) (w dw) c -> (h w) (dh dw c)',
|
| 322 |
+
dh=base_image_input_d,
|
| 323 |
+
dw=base_image_input_d,
|
| 324 |
+
h=image_base_patch_h,
|
| 325 |
+
w=image_base_patch_w
|
| 326 |
+
)
|
| 327 |
+
patches = np.concatenate([np.expand_dims(resized, 0), patches], 0)
|
| 328 |
+
|
| 329 |
+
# Global image goes first, so the order of patches in previous crops gets increased
|
| 330 |
+
patch_ordering = np.where(
|
| 331 |
+
patch_ordering >= 0,
|
| 332 |
+
patch_ordering + tokens_per_image,
|
| 333 |
+
-1
|
| 334 |
+
)
|
| 335 |
+
patch_ordering = np.concatenate([np.arange(0, tokens_per_image), patch_ordering], 0)
|
| 336 |
+
per_row = np.full(
|
| 337 |
+
(image_token_length_w,),
|
| 338 |
+
image_patch_token_id,
|
| 339 |
+
)
|
| 340 |
+
per_row = np.concatenate([per_row, [image_col_token_id]], 0)
|
| 341 |
+
extra_tokens = np.tile(per_row, [image_token_length_h])
|
| 342 |
+
joint = [
|
| 343 |
+
[image_start_token_id],
|
| 344 |
+
extra_tokens,
|
| 345 |
+
[image_end_token_id],
|
| 346 |
+
] + joint
|
| 347 |
+
|
| 348 |
+
joint = np.concatenate(joint, 0)
|
| 349 |
+
img_mask = np.pad(img_mask, [[0, 1], [0, 0]], constant_values=-1)
|
| 350 |
+
return patches, joint, patch_ordering, img_mask
|
| 351 |
+
|
| 352 |
+
def build_image_input_idx(
|
| 353 |
+
self,
|
| 354 |
+
image_tokens: np.ndarray,
|
| 355 |
+
patch_order: np.ndarray,
|
| 356 |
+
image_patch_token_id: int,
|
| 357 |
+
no_image: Optional[bool] = None,
|
| 358 |
+
image_token_length_w: Optional[int] = None,
|
| 359 |
+
image_token_length_h: Optional[int] = None,
|
| 360 |
+
):
|
| 361 |
+
"""Converts `patch_order` into a mapping of token_id -> patch_id"""
|
| 362 |
+
|
| 363 |
+
tokens_per_image = image_token_length_w * image_token_length_h
|
| 364 |
+
if no_image is not None and no_image:
|
| 365 |
+
return np.zeros((0, tokens_per_image), np.int32)
|
| 366 |
+
|
| 367 |
+
# Indices to insert the patches
|
| 368 |
+
image_input_idx = image_tokens == image_patch_token_id
|
| 369 |
+
image_input_idx = np.nonzero(image_input_idx)[0].astype(np.int32)
|
| 370 |
+
|
| 371 |
+
if patch_order is not None:
|
| 372 |
+
n_tokens = image_input_idx.shape[0]
|
| 373 |
+
patch_order = np.reshape(patch_order, [-1])
|
| 374 |
+
n_patches = patch_order.shape[0]
|
| 375 |
+
|
| 376 |
+
valid = patch_order >= 0
|
| 377 |
+
n_valid_patches = valid.sum()
|
| 378 |
+
assert len(image_input_idx) == n_valid_patches
|
| 379 |
+
|
| 380 |
+
sorted_patch_ixs = np.zeros([n_tokens], np.int32)
|
| 381 |
+
sorted_patch_ixs[patch_order[valid]] = np.arange(n_valid_patches, dtype=np.int32)
|
| 382 |
+
|
| 383 |
+
# Project the inverted mapping into same sparse structure
|
| 384 |
+
sorted_patch_ixs_ex = np.full(np.shape(patch_order), -1)
|
| 385 |
+
sorted_patch_ixs_ex[valid] = sorted_patch_ixs
|
| 386 |
+
|
| 387 |
+
# Do the gather and then re-masked outputs that were masked in `sorted_patch_ixs`
|
| 388 |
+
valid = (sorted_patch_ixs_ex >= 0).astype(np.int32)
|
| 389 |
+
image_input_idx = image_input_idx[sorted_patch_ixs_ex*valid]
|
| 390 |
+
image_input_idx = image_input_idx*valid - 100*(1 - valid)
|
| 391 |
+
image_input_idx = np.reshape(image_input_idx, [-1, tokens_per_image])
|
| 392 |
+
return image_input_idx
|
| 393 |
+
|
| 394 |
+
def preprocess(
|
| 395 |
+
self,
|
| 396 |
+
image: np.ndarray,
|
| 397 |
+
image_patch_token_id: int,
|
| 398 |
+
image_col_token_id: int,
|
| 399 |
+
image_start_token_id: int,
|
| 400 |
+
image_end_token_id: int,
|
| 401 |
+
max_crops: Optional[int] = None,
|
| 402 |
+
overlap_margins: Optional[List[int]] = None,
|
| 403 |
+
base_image_input_size: Optional[Union[int, List[int]]] = None,
|
| 404 |
+
image_token_length_w: Optional[int] = None,
|
| 405 |
+
image_token_length_h: Optional[int] = None,
|
| 406 |
+
image_patch_size: Optional[int] = None,
|
| 407 |
+
**kwargs,
|
| 408 |
+
):
|
| 409 |
+
"""Preprocesses an image
|
| 410 |
+
|
| 411 |
+
Returns:
|
| 412 |
+
crops: (n_crops, n_patches, patch_dim) individual crops, `n_crops` might
|
| 413 |
+
change between images but the other dimension are fixed
|
| 414 |
+
tokens: (n_tokens,) int32 tokens, pad tokens indicate where to insert the
|
| 415 |
+
patch features, might include other special tokens as well
|
| 416 |
+
image_idx: (n_crops, n_patches) index in `tokens` to put the patch features from the
|
| 417 |
+
crops after pooling, negative values indicates patches features to exclude
|
| 418 |
+
padding_mask: (n_crops, n_patches) what percent of each crop is padding, can be None
|
| 419 |
+
if the image mask is not being used.
|
| 420 |
+
"""
|
| 421 |
+
|
| 422 |
+
max_crops = max_crops or self.max_crops
|
| 423 |
+
overlap_margins = overlap_margins or self.overlap_margins
|
| 424 |
+
base_image_input_size = base_image_input_size or self.base_image_input_size
|
| 425 |
+
image_token_length_w = image_token_length_w or self.image_token_length_w
|
| 426 |
+
image_token_length_h = image_token_length_h or self.image_token_length_h
|
| 427 |
+
image_patch_size = image_patch_size or self.image_patch_size
|
| 428 |
+
|
| 429 |
+
crops, image_tokens, patch_ordering, img_mask = self.image_to_patches_and_tokens(
|
| 430 |
+
image,
|
| 431 |
+
image_patch_token_id,
|
| 432 |
+
image_col_token_id,
|
| 433 |
+
image_start_token_id,
|
| 434 |
+
image_end_token_id,
|
| 435 |
+
max_crops,
|
| 436 |
+
overlap_margins,
|
| 437 |
+
base_image_input_size,
|
| 438 |
+
image_token_length_w,
|
| 439 |
+
image_token_length_h,
|
| 440 |
+
image_patch_size,
|
| 441 |
+
)
|
| 442 |
+
patch_idx = self.build_image_input_idx(
|
| 443 |
+
image_tokens,
|
| 444 |
+
patch_ordering,
|
| 445 |
+
image_patch_token_id,
|
| 446 |
+
image_token_length_w=image_token_length_w,
|
| 447 |
+
image_token_length_h=image_token_length_h,
|
| 448 |
+
)
|
| 449 |
+
return crops, image_tokens, patch_idx, img_mask
|
| 450 |
+
|
| 451 |
+
def multimodal_preprocess(
|
| 452 |
+
self,
|
| 453 |
+
images: np.ndarray,
|
| 454 |
+
tokens: List[int],
|
| 455 |
+
image_idx: np.ndarray,
|
| 456 |
+
sequence_length: int,
|
| 457 |
+
image_patch_token_id: int,
|
| 458 |
+
image_col_token_id: int,
|
| 459 |
+
image_start_token_id: int,
|
| 460 |
+
image_end_token_id: int,
|
| 461 |
+
**kwargs,
|
| 462 |
+
):
|
| 463 |
+
"""Merge images and text tokens into multi-modal features for the model
|
| 464 |
+
|
| 465 |
+
:param images: images to use as input
|
| 466 |
+
:param tokens: input text tokens
|
| 467 |
+
:param image_idx: where to insert the images into `tokens`
|
| 468 |
+
:params image_patch_token_id: id to use of tokens that will contain image features
|
| 469 |
+
:params image_col_token_id: token id for image column special tokens
|
| 470 |
+
:params image_start_token_id: token id for image start special tokens
|
| 471 |
+
:params image_end_token_id: token id for image end special tokens
|
| 472 |
+
:params kwargs: override preprocessor default args
|
| 473 |
+
"""
|
| 474 |
+
max_total_crops = kwargs.get("max_crops") or self.max_crops
|
| 475 |
+
image_token_length_w = kwargs.get("image_token_length_w") or self.image_token_length_w
|
| 476 |
+
image_token_length_h = kwargs.get("image_token_length_h") or self.image_token_length_h
|
| 477 |
+
image_patch_size = kwargs.get("image_patch_size") or self.image_patch_size
|
| 478 |
+
base_image_input_size = kwargs.get("base_image_input_size") or self.base_image_input_size
|
| 479 |
+
image_num_patch = (
|
| 480 |
+
base_image_input_size[0] // image_patch_size,
|
| 481 |
+
base_image_input_size[1] // image_patch_size,
|
| 482 |
+
)
|
| 483 |
+
image_padding_mask = kwargs.get("image_padding_mask") or self.image_padding_mask
|
| 484 |
+
|
| 485 |
+
tokens_per_image = image_token_length_w * image_token_length_h
|
| 486 |
+
n_pixels = image_patch_size * image_patch_size * 3
|
| 487 |
+
n_patches = image_num_patch[0] * image_num_patch[1]
|
| 488 |
+
|
| 489 |
+
if images is None:
|
| 490 |
+
return {
|
| 491 |
+
"input_ids": tokens,
|
| 492 |
+
"images": None,
|
| 493 |
+
"image_input_idx": None
|
| 494 |
+
}
|
| 495 |
+
else:
|
| 496 |
+
n = len(images)
|
| 497 |
+
all_crops = []
|
| 498 |
+
all_image_idx = []
|
| 499 |
+
out_tokens = []
|
| 500 |
+
all_crop_masks = []
|
| 501 |
+
|
| 502 |
+
for ix in range(n):
|
| 503 |
+
token_ix = image_idx[ix]
|
| 504 |
+
crops, image_tokens, patch_idx, img_mask = self.preprocess(
|
| 505 |
+
images[ix],
|
| 506 |
+
image_patch_token_id,
|
| 507 |
+
image_col_token_id,
|
| 508 |
+
image_start_token_id,
|
| 509 |
+
image_end_token_id,
|
| 510 |
+
**kwargs,
|
| 511 |
+
)
|
| 512 |
+
|
| 513 |
+
if token_ix == -1: # -1 is an image inserted at the very start
|
| 514 |
+
start = 0
|
| 515 |
+
token_ix = 0
|
| 516 |
+
end = 0
|
| 517 |
+
else:
|
| 518 |
+
start = 0 if ix == 0 else image_idx[ix-1] + 1
|
| 519 |
+
end = token_ix + 1
|
| 520 |
+
|
| 521 |
+
all_image_idx.append(patch_idx + token_ix)
|
| 522 |
+
all_crops.append(crops)
|
| 523 |
+
out_tokens.append(tokens[start:token_ix])
|
| 524 |
+
out_tokens.append(image_tokens)
|
| 525 |
+
if ix == (n - 1):
|
| 526 |
+
out_tokens.append(tokens[end:])
|
| 527 |
+
if image_padding_mask:
|
| 528 |
+
all_crop_masks.append(img_mask)
|
| 529 |
+
|
| 530 |
+
input_ids = np.concatenate(out_tokens, 0)
|
| 531 |
+
images = np.concatenate(all_crops, 0)
|
| 532 |
+
image_input_idx = np.concatenate(all_image_idx, 0)
|
| 533 |
+
if image_padding_mask:
|
| 534 |
+
image_masks = np.concatenate(all_crop_masks, 0)
|
| 535 |
+
else:
|
| 536 |
+
image_masks = None
|
| 537 |
+
|
| 538 |
+
out = {
|
| 539 |
+
"input_ids": input_ids,
|
| 540 |
+
"images": images,
|
| 541 |
+
"image_input_idx": image_input_idx
|
| 542 |
+
}
|
| 543 |
+
if image_masks is not None:
|
| 544 |
+
out["image_masks"] = image_masks
|
| 545 |
+
return out
|
| 546 |
+
|
| 547 |
+
|
| 548 |
+
MolmoImageProcessor.register_for_auto_class()
|