Spaces:
Running
on
Zero
Running
on
Zero
File size: 7,780 Bytes
9b33fca |
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 |
# pylint: disable=duplicate-code
"""COCO data loading config for YOLOX object detection."""
from __future__ import annotations
from collections.abc import Sequence
from ml_collections import ConfigDict
from vis4d.config import class_config
from vis4d.config.typing import DataConfig
from vis4d.data.const import CommonKeys as K
from vis4d.data.data_pipe import DataPipe, MultiSampleDataPipe
from vis4d.data.datasets.coco import COCO
from vis4d.data.io import DataBackend
from vis4d.data.loader import build_train_dataloader, default_collate
from vis4d.data.transforms.affine import (
AffineBoxes2D,
AffineImages,
GenAffineParameters,
)
from vis4d.data.transforms.base import RandomApply, compose
from vis4d.data.transforms.flip import FlipBoxes2D, FlipImages
from vis4d.data.transforms.mixup import (
GenMixupParameters,
MixupBoxes2D,
MixupImages,
)
from vis4d.data.transforms.mosaic import (
GenMosaicParameters,
MosaicBoxes2D,
MosaicImages,
)
from vis4d.data.transforms.pad import PadImages
from vis4d.data.transforms.photometric import RandomHSV
from vis4d.data.transforms.post_process import PostProcessBoxes2D
from vis4d.data.transforms.resize import (
GenResizeParameters,
ResizeBoxes2D,
ResizeImages,
)
from vis4d.data.transforms.to_tensor import ToTensor
from vis4d.engine.connectors import data_key, pred_key
from vis4d.zoo.base import get_inference_dataloaders_cfg
from vis4d.zoo.base.callable import get_callable_cfg
CONN_COCO_BBOX_EVAL = {
"coco_image_id": data_key(K.sample_names),
"pred_boxes": pred_key("boxes"),
"pred_scores": pred_key("scores"),
"pred_classes": pred_key("class_ids"),
}
CONN_COCO_MASK_EVAL = {
"coco_image_id": data_key(K.sample_names),
"pred_boxes": pred_key("boxes.boxes"),
"pred_scores": pred_key("boxes.scores"),
"pred_classes": pred_key("boxes.class_ids"),
"pred_masks": pred_key("masks"),
}
def get_train_dataloader(
data_root: str,
split: str,
keys_to_load: Sequence[str],
data_backend: None | DataBackend,
image_size: tuple[int, int],
scaling_ratio_range: tuple[float, float],
use_mixup: bool,
samples_per_gpu: int,
workers_per_gpu: int,
) -> ConfigDict:
"""Get the default train dataloader for COCO detection."""
# Train Dataset
train_dataset_cfg = class_config(
COCO,
keys_to_load=keys_to_load,
data_root=data_root,
split=split,
remove_empty=False,
image_channel_mode="BGR",
data_backend=data_backend,
)
# Train Preprocessing
preprocess_transforms = [
[
class_config(GenMosaicParameters, out_shape=image_size),
class_config(MosaicImages, imresize_backend="cv2"),
class_config(MosaicBoxes2D),
]
]
preprocess_transforms += [
[
class_config(
GenAffineParameters,
scaling_ratio_range=scaling_ratio_range,
border=(-image_size[0] // 2, -image_size[1] // 2),
),
class_config(AffineImages, as_int=True),
class_config(AffineBoxes2D),
]
]
if use_mixup:
preprocess_transforms += [
[
class_config(
GenMixupParameters,
out_shape=image_size,
mixup_ratio_dist="const",
scale_range=(0.8, 1.6),
pad_value=114.0,
),
class_config(MixupImages, imresize_backend="cv2"),
class_config(MixupBoxes2D),
]
]
preprocess_transforms.append(
[class_config(PostProcessBoxes2D, min_area=1.0)]
)
train_batchprocess_cfg = class_config(
compose,
transforms=[
class_config(RandomHSV, same_on_batch=False),
class_config(
RandomApply,
transforms=[
class_config(FlipImages),
class_config(FlipBoxes2D),
],
probability=0.5,
same_on_batch=False,
),
class_config(
GenResizeParameters,
shape=image_size,
keep_ratio=True,
same_on_batch=False,
),
class_config(ResizeImages, imresize_backend="cv2"),
class_config(ResizeBoxes2D),
class_config(PadImages, value=114.0, pad2square=True),
class_config(ToTensor),
],
)
return class_config(
build_train_dataloader,
dataset=class_config(
MultiSampleDataPipe,
datasets=train_dataset_cfg,
preprocess_fn=preprocess_transforms,
),
samples_per_gpu=samples_per_gpu,
workers_per_gpu=workers_per_gpu,
batchprocess_fn=train_batchprocess_cfg,
collate_fn=get_callable_cfg(default_collate),
pin_memory=True,
shuffle=True,
)
def get_test_dataloader(
data_root: str,
split: str,
keys_to_load: Sequence[str],
data_backend: None | DataBackend,
image_size: tuple[int, int],
samples_per_gpu: int,
workers_per_gpu: int,
) -> ConfigDict:
"""Get the default test dataloader for COCO detection."""
# Test Dataset
test_dataset = class_config(
COCO,
keys_to_load=keys_to_load,
data_root=data_root,
split=split,
image_channel_mode="BGR",
data_backend=data_backend,
)
# Test Preprocessing
preprocess_transforms = [
class_config(GenResizeParameters, shape=image_size, keep_ratio=True),
class_config(ResizeImages, imresize_backend="cv2"),
]
test_preprocess_cfg = class_config(
compose, transforms=preprocess_transforms
)
test_batchprocess_cfg = class_config(
compose,
transforms=[
class_config(PadImages, value=114.0, pad2square=True),
class_config(ToTensor),
],
)
# Test Dataset Config
test_dataset_cfg = class_config(
DataPipe, datasets=test_dataset, preprocess_fn=test_preprocess_cfg
)
return get_inference_dataloaders_cfg(
datasets_cfg=test_dataset_cfg,
batchprocess_cfg=test_batchprocess_cfg,
samples_per_gpu=samples_per_gpu,
workers_per_gpu=workers_per_gpu,
)
def get_coco_yolox_cfg(
data_root: str = "data/coco",
train_split: str = "train2017",
train_keys_to_load: Sequence[str] = (
K.images,
K.boxes2d,
K.boxes2d_classes,
),
test_split: str = "val2017",
test_keys_to_load: Sequence[str] = (K.images, K.original_images),
data_backend: None | ConfigDict = None,
train_image_size: tuple[int, int] = (640, 640),
scaling_ratio_range: tuple[float, float] = (0.1, 2.0),
use_mixup: bool = True,
test_image_size: tuple[int, int] = (640, 640),
samples_per_gpu: int = 2,
workers_per_gpu: int = 2,
) -> DataConfig:
"""Get the default config for COCO detection."""
data = DataConfig()
data.train_dataloader = get_train_dataloader(
data_root=data_root,
split=train_split,
keys_to_load=train_keys_to_load,
data_backend=data_backend,
image_size=train_image_size,
scaling_ratio_range=scaling_ratio_range,
use_mixup=use_mixup,
samples_per_gpu=samples_per_gpu,
workers_per_gpu=workers_per_gpu,
)
data.test_dataloader = get_test_dataloader(
data_root=data_root,
split=test_split,
keys_to_load=test_keys_to_load,
data_backend=data_backend,
image_size=test_image_size,
samples_per_gpu=1,
workers_per_gpu=workers_per_gpu,
)
return data
|