Spaces:
Runtime error
Runtime error
File size: 9,162 Bytes
f549064 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 |
# Copyright (c) OpenMMLab. All rights reserved.
from collections import defaultdict
from collections.abc import Sequence
from functools import partial
import numpy as np
import torch
from mmcv.transforms import BaseTransform
from mmengine.utils import is_str
from PIL import Image
from mmcls.registry import TRANSFORMS
from mmcls.structures import ClsDataSample, MultiTaskDataSample
def to_tensor(data):
"""Convert objects of various python types to :obj:`torch.Tensor`.
Supported types are: :class:`numpy.ndarray`, :class:`torch.Tensor`,
:class:`Sequence`, :class:`int` and :class:`float`.
"""
if isinstance(data, torch.Tensor):
return data
elif isinstance(data, np.ndarray):
return torch.from_numpy(data)
elif isinstance(data, Sequence) and not is_str(data):
return torch.tensor(data)
elif isinstance(data, int):
return torch.LongTensor([data])
elif isinstance(data, float):
return torch.FloatTensor([data])
else:
raise TypeError(
f'Type {type(data)} cannot be converted to tensor.'
'Supported types are: `numpy.ndarray`, `torch.Tensor`, '
'`Sequence`, `int` and `float`')
@TRANSFORMS.register_module()
class PackClsInputs(BaseTransform):
"""Pack the inputs data for the classification.
**Required Keys:**
- img
- gt_label (optional)
- ``*meta_keys`` (optional)
**Deleted Keys:**
All keys in the dict.
**Added Keys:**
- inputs (:obj:`torch.Tensor`): The forward data of models.
- data_samples (:obj:`~mmcls.structures.ClsDataSample`): The annotation
info of the sample.
Args:
meta_keys (Sequence[str]): The meta keys to be saved in the
``metainfo`` of the packed ``data_samples``.
Defaults to a tuple includes keys:
- ``sample_idx``: The id of the image sample.
- ``img_path``: The path to the image file.
- ``ori_shape``: The original shape of the image as a tuple (H, W).
- ``img_shape``: The shape of the image after the pipeline as a
tuple (H, W).
- ``scale_factor``: The scale factor between the resized image and
the original image.
- ``flip``: A boolean indicating if image flip transform was used.
- ``flip_direction``: The flipping direction.
"""
def __init__(self,
meta_keys=('sample_idx', 'img_path', 'ori_shape', 'img_shape',
'scale_factor', 'flip', 'flip_direction')):
self.meta_keys = meta_keys
def transform(self, results: dict) -> dict:
"""Method to pack the input data."""
packed_results = dict()
if 'img' in results:
img = results['img']
if len(img.shape) < 3:
img = np.expand_dims(img, -1)
img = np.ascontiguousarray(img.transpose(2, 0, 1))
packed_results['inputs'] = to_tensor(img)
data_sample = ClsDataSample()
if 'gt_label' in results:
gt_label = results['gt_label']
data_sample.set_gt_label(gt_label)
img_meta = {k: results[k] for k in self.meta_keys if k in results}
data_sample.set_metainfo(img_meta)
packed_results['data_samples'] = data_sample
return packed_results
def __repr__(self) -> str:
repr_str = self.__class__.__name__
repr_str += f'(meta_keys={self.meta_keys})'
return repr_str
@TRANSFORMS.register_module()
class PackMultiTaskInputs(BaseTransform):
"""Convert all image labels of multi-task dataset to a dict of tensor.
Args:
tasks (List[str]): The task names defined in the dataset.
meta_keys(Sequence[str]): The meta keys to be saved in the
``metainfo`` of the packed ``data_samples``.
Defaults to a tuple includes keys:
- ``sample_idx``: The id of the image sample.
- ``img_path``: The path to the image file.
- ``ori_shape``: The original shape of the image as a tuple (H, W).
- ``img_shape``: The shape of the image after the pipeline as a
tuple (H, W).
- ``scale_factor``: The scale factor between the resized image and
the original image.
- ``flip``: A boolean indicating if image flip transform was used.
- ``flip_direction``: The flipping direction.
"""
def __init__(self,
task_handlers=dict(),
multi_task_fields=('gt_label', ),
meta_keys=('sample_idx', 'img_path', 'ori_shape', 'img_shape',
'scale_factor', 'flip', 'flip_direction')):
self.multi_task_fields = multi_task_fields
self.meta_keys = meta_keys
self.task_handlers = defaultdict(
partial(PackClsInputs, meta_keys=meta_keys))
for task_name, task_handler in task_handlers.items():
self.task_handlers[task_name] = TRANSFORMS.build(
dict(type=task_handler, meta_keys=meta_keys))
def transform(self, results: dict) -> dict:
"""Method to pack the input data.
result = {'img_path': 'a.png', 'gt_label': {'task1': 1, 'task3': 3},
'img': array([[[ 0, 0, 0])
"""
packed_results = dict()
results = results.copy()
if 'img' in results:
img = results.pop('img')
if len(img.shape) < 3:
img = np.expand_dims(img, -1)
img = np.ascontiguousarray(img.transpose(2, 0, 1))
packed_results['inputs'] = to_tensor(img)
task_results = defaultdict(dict)
for field in self.multi_task_fields:
if field in results:
value = results.pop(field)
for k, v in value.items():
task_results[k].update({field: v})
data_sample = MultiTaskDataSample()
for task_name, task_result in task_results.items():
task_handler = self.task_handlers[task_name]
task_pack_result = task_handler({**results, **task_result})
data_sample.set_field(task_pack_result['data_samples'], task_name)
packed_results['data_samples'] = data_sample
return packed_results
def __repr__(self):
repr = self.__class__.__name__
task_handlers = {
name: handler.__class__.__name__
for name, handler in self.task_handlers.items()
}
repr += f'(task_handlers={task_handlers}, '
repr += f'multi_task_fields={self.multi_task_fields}, '
repr += f'meta_keys={self.meta_keys})'
return repr
@TRANSFORMS.register_module()
class Transpose(BaseTransform):
"""Transpose numpy array.
**Required Keys:**
- ``*keys``
**Modified Keys:**
- ``*keys``
Args:
keys (List[str]): The fields to convert to tensor.
order (List[int]): The output dimensions order.
"""
def __init__(self, keys, order):
self.keys = keys
self.order = order
def transform(self, results):
"""Method to transpose array."""
for key in self.keys:
results[key] = results[key].transpose(self.order)
return results
def __repr__(self):
return self.__class__.__name__ + \
f'(keys={self.keys}, order={self.order})'
@TRANSFORMS.register_module()
class ToPIL(BaseTransform):
"""Convert the image from OpenCV format to :obj:`PIL.Image.Image`.
**Required Keys:**
- img
**Modified Keys:**
- img
"""
def transform(self, results):
"""Method to convert images to :obj:`PIL.Image.Image`."""
results['img'] = Image.fromarray(results['img'])
return results
@TRANSFORMS.register_module()
class ToNumpy(BaseTransform):
"""Convert object to :obj:`numpy.ndarray`.
**Required Keys:**
- ``*keys**``
**Modified Keys:**
- ``*keys**``
Args:
dtype (str, optional): The dtype of the converted numpy array.
Defaults to None.
"""
def __init__(self, keys, dtype=None):
self.keys = keys
self.dtype = dtype
def transform(self, results):
"""Method to convert object to :obj:`numpy.ndarray`."""
for key in self.keys:
results[key] = np.array(results[key], dtype=self.dtype)
return results
def __repr__(self):
return self.__class__.__name__ + \
f'(keys={self.keys}, dtype={self.dtype})'
@TRANSFORMS.register_module()
class Collect(BaseTransform):
"""Collect and only reserve the specified fields.
**Required Keys:**
- ``*keys``
**Deleted Keys:**
All keys except those in the argument ``*keys``.
Args:
keys (Sequence[str]): The keys of the fields to be collected.
"""
def __init__(self, keys):
self.keys = keys
def transform(self, results):
data = {}
for key in self.keys:
data[key] = results[key]
return data
def __repr__(self):
return self.__class__.__name__ + f'(keys={self.keys})'
|