File size: 7,531 Bytes
bdec3d7 |
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 |
# --------------------------------------------------------
# X-Decoder -- Generalized Decoding for Pixel, Image, and Language
# Copyright (c) 2022 Microsoft
# Licensed under The MIT License [see LICENSE for details]
# Modified by Xueyan Zou ([email protected])
# --------------------------------------------------------
# Copyright (c) Facebook, Inc. and its affiliates.
import copy
import random
import scipy.io
import numpy as np
import torch
from PIL import Image
from torchvision import transforms
from pycocotools import mask
from detectron2.data import detection_utils as utils
from detectron2.data import transforms as T
from modeling.utils import configurable
__all__ = ["RefCOCODatasetMapper"]
def build_transform_gen(cfg, is_train):
"""
Create a list of default :class:`Augmentation` from config.
Now it includes resizing and flipping.
Returns:
list[Augmentation]
"""
assert is_train, "Only support training augmentation"
cfg_input = cfg['INPUT']
image_size = cfg_input['IMAGE_SIZE']
min_scale = cfg_input['MIN_SCALE']
max_scale = cfg_input['MAX_SCALE']
augmentation = []
if cfg_input['RANDOM_FLIP'] != "none":
augmentation.append(
T.RandomFlip(
horizontal=cfg_input['RANDOM_FLIP'] == "horizontal",
vertical=cfg_input['RANDOM_FLIP'] == "vertical",
)
)
augmentation.extend([
T.ResizeScale(
min_scale=min_scale, max_scale=max_scale, target_height=image_size, target_width=image_size
),
T.FixedSizeCrop(crop_size=(image_size, image_size)),
])
return augmentation
def build_transform_gen_se(cfg, is_train):
min_scale = cfg['INPUT']['MIN_SIZE_TEST']
max_scale = cfg['INPUT']['MAX_SIZE_TEST']
augmentation = []
augmentation.extend([
T.ResizeShortestEdge(
min_scale, max_size=max_scale
),
])
return augmentation
# This is specifically designed for the COCO dataset.
class RefCOCODatasetMapper:
"""
A callable which takes a dataset dict in Detectron2 Dataset format,
and map it into a format used by MaskFormer.
This dataset mapper applies the same transformation as DETR for COCO panoptic segmentation.
The callable currently does the following:
1. Read the image from "file_name"
2. Applies geometric transforms to the image and annotation
3. Find and applies suitable cropping to the image and annotation
4. Prepare image and annotation to Tensors
"""
@configurable
def __init__(
self,
is_train=True,
tfm_gens=None,
image_format=None,
min_size_test=None,
max_size_test=None,
mean=None,
std=None,
):
"""
NOTE: this interface is experimental.
Args:
is_train: for training or inference
augmentations: a list of augmentations or deterministic transforms to apply
tfm_gens: data augmentation
image_format: an image format supported by :func:`detection_utils.read_image`.
"""
self.tfm_gens = tfm_gens
self.img_format = image_format
self.is_train = is_train
self.min_size_test = min_size_test
self.max_size_test = max_size_test
self.pixel_mean = torch.tensor(mean)[:,None,None]
self.pixel_std = torch.tensor(std)[:,None,None]
# t = []
# t.append(T.ResizeShortestEdge(min_size_test, max_size=max_size_test))
# self.transform = transforms.Compose(t)
@classmethod
def from_config(cls, cfg, is_train=True):
# Build augmentation
if is_train:
tfm_gens = build_transform_gen(cfg, is_train)
else:
tfm_gens = build_transform_gen_se(cfg, is_train)
ret = {
"is_train": is_train,
"tfm_gens": tfm_gens,
"image_format": cfg['INPUT'].get('FORMAT', 'RGB'),
"min_size_test": cfg['INPUT']['MIN_SIZE_TEST'],
"max_size_test": cfg['INPUT']['MAX_SIZE_TEST'],
"mean": cfg['INPUT']['PIXEL_MEAN'],
"std": cfg['INPUT']['PIXEL_STD'],
}
return ret
def __call__(self, dataset_dict):
"""
Args:
dataset_dict (dict): Metadata of one image, in Detectron2 Dataset format.
Returns:
dict: a format that builtin models in detectron2 accept
"""
dataset_dict = copy.deepcopy(dataset_dict) # it will be modified by code below
file_name = dataset_dict['file_name']
if self.is_train == False:
image = utils.read_image(file_name, format=self.img_format)
utils.check_image_size(dataset_dict, image)
image, _ = T.apply_transform_gens(self.tfm_gens, image)
dataset_dict["image"] = torch.as_tensor(np.ascontiguousarray(image.transpose(2, 0, 1)))
grounding_anno = dataset_dict['grounding_info']
assert len(grounding_anno) > 0
masks_grd = []
texts_grd = []
boxes_grd = []
for ann in grounding_anno:
rle = mask.frPyObjects(
ann['segmentation'], dataset_dict['height'], dataset_dict['width'])
m = mask.decode(rle)
# sometimes there are multiple binary map (corresponding to multiple segs)
m = np.sum(m, axis=2)
m = m.astype(np.uint8) # convert to np.uint8
masks_grd += [m]
texts_grd.append([x['raw'].lower() for x in ann['sentences']])
boxes_grd.append(ann['bbox']) # xywh
masks_grd = torch.from_numpy(np.stack(masks_grd))
boxes_grd = torch.tensor(boxes_grd)
groundings = {'masks': masks_grd, 'texts': texts_grd, 'boxes': boxes_grd}
dataset_dict["groundings"] = groundings
else:
image = utils.read_image(dataset_dict["file_name"], format=self.img_format)
utils.check_image_size(dataset_dict, image)
image, transforms = T.apply_transform_gens(self.tfm_gens, image)
image_shape = image.shape[:2] # h, w
dataset_dict["image"] = torch.as_tensor(np.ascontiguousarray(image.transpose(2, 0, 1)))
grounding_anno = dataset_dict['grounding_info']
assert len(grounding_anno) > 0
masks_grd = []
texts_grd = []
boxes_grd = []
hash_grd = []
for ann in grounding_anno:
rle = mask.frPyObjects(
ann['segmentation'], dataset_dict['height'], dataset_dict['width'])
m = mask.decode(rle)
# sometimes there are multiple binary map (corresponding to multiple segs)
m = np.sum(m, axis=2)
m = m.astype(np.uint8) # convert to np.uint8
m = transforms.apply_segmentation(m[:,:,None])[:,:,0]
masks_grd += [m]
rand_id = random.randint(0, len(ann['sentences'])-1)
texts_grd.append(ann['sentences'][rand_id]['raw'].lower())
hash_grd.append(hash(ann['sentences'][rand_id]['raw'].lower()))
masks_grd = torch.from_numpy(np.stack(masks_grd))
boxes_grd = torch.tensor(boxes_grd)
groundings = {'masks': masks_grd, 'texts': texts_grd, 'hash': hash_grd, 'mode': 'text'}
dataset_dict["groundings"] = groundings
return dataset_dict |