mirrors/mmengine
1
0
Fork 0
mirror of https://github.com/open-mmlab/mmengine.git synced 2025-06-03 21:54:44 +08:00
Code Issues Projects Releases Wiki Activity

[Fix] fix random may generate same value (#85)

Browse Source
This commit is contained in:
liukuikun 2022-03-05 15:02:54 +08:00 committed by GitHub
parent 43dcc5648f
commit 11b38b12d6
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 98 additions and 98 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

69
mmengine/data/base_data_element.py
View File

@ -25,27 +25,26 @@ class BaseDataElement:
types of ground truth labels or predictions.
They are used as interfaces between different commopenets.
The attributes in ``BaseDataElement`` are divided into two parts,
the ``metainfo`` and the ``data`` respectively.
- ``metainfo``: Usually contains the
information about the image such as filename,
image_shape, pad_shape, etc. The attributes can be accessed or
modified by dict-like or object-like operations, such as
``.``(for data access and modification) , ``in``, ``del``,
``pop(str)``, ``get(str)``, ``metainfo_keys()``,
``metainfo_values()``, ``metainfo_items()``, ``set_metainfo()``(for
set or change key-value pairs in metainfo).
- ``metainfo``: Usually contains the
information about the image such as filename,
image_shape, pad_shape, etc. The attributes can be accessed or
modified by dict-like or object-like operations, such as
``.``(for data access and modification) , ``in``, ``del``,
``pop(str)``, ``get(str)``, ``metainfo_keys()``,
``metainfo_values()``, ``metainfo_items()``, ``set_metainfo()``(for
set or change key-value pairs in metainfo).
- ``data``: Annotations or model predictions are
stored. The attributes can be accessed or modified by
dict-like or object-like operations, such as
``.`` , ``in``, ``del``, ``pop(str)`` ``get(str)``, ``data_keys()``,
``data_values()``, ``data_items()``. Users can also apply tensor-like
methods to all obj:``torch.Tensor`` in the ``data_fileds``,
such as ``.cuda()``, ``.cpu()``, ``.numpy()``, , ``.to()``
``to_tensor()``, ``.detach()``, ``.numpy()``
- ``data``: Annotations or model predictions are
stored. The attributes can be accessed or modified by
dict-like or object-like operations, such as
``.`` , ``in``, ``del``, ``pop(str)`` ``get(str)``, ``data_keys()``,
``data_values()``, ``data_items()``. Users can also apply tensor-like
methods to all obj:``torch.Tensor`` in the ``data_fileds``,
such as ``.cuda()``, ``.cpu()``, ``.numpy()``, , ``.to()``
``to_tensor()``, ``.detach()``, ``.numpy()``
Args:
meta_info (dict, optional): A dict contains the meta information
@ -57,24 +56,24 @@ class BaseDataElement:
Examples:
>>> from mmengine.data import BaseDataElement
>>> gt_instances = BaseDataElement()
>>> bboxes = torch.rand((5, 4))
>>> scores = torch.rand((5,))
>>> img_id = 0
>>> img_shape = (800, 1333)
>>> gt_instances = BaseDataElement(
metainfo=dict(img_id=img_id, img_shape=img_shape),
data=dict(bboxes=bboxes, scores=scores))
... metainfo=dict(img_id=img_id, img_shape=img_shape),
... data=dict(bboxes=bboxes, scores=scores))
>>> gt_instances = BaseDataElement(dict(img_id=img_id,
img_shape=(H, W)))
# new
... img_shape=(H, W)))
>>> # new
>>> gt_instances1 = gt_instance.new(
metainfo=dict(img_id=1, img_shape=(640, 640)),
data=dict(bboxes=torch.rand((5, 4)),
scores=torch.rand((5,))))
... metainfo=dict(img_id=1, img_shape=(640, 640)),
... data=dict(bboxes=torch.rand((5, 4)),
... scores=torch.rand((5,))))
>>> gt_instances2 = gt_instances1.new()
# add and process property
>>> # add and process property
>>> gt_instances = BaseDataElement()
>>> gt_instances.set_metainfo(dict(img_id=9, img_shape=(100, 100))
>>> assert 'img_shape' in gt_instances.metainfo_keys()
@ -82,14 +81,12 @@ class BaseDataElement:
>>> assert 'img_shape' not in gt_instances.data_keys()
>>> assert 'img_shape' in gt_instances.keys()
>>> print(gt_instances.img_shape)
>>> gt_instances.scores = torch.rand((5,))
>>> assert 'scores' in gt_instances.data_keys()
>>> assert 'scores' in gt_instances
>>> assert 'scores' in gt_instances.keys()
>>> assert 'scores' not in gt_instances.metainfo_keys()
>>> print(gt_instances.scores)
>>> gt_instances.bboxes = torch.rand((5, 4))
>>> assert 'bboxes' in gt_instances.data_keys()
>>> assert 'bboxes' in gt_instances
@ -97,14 +94,14 @@ class BaseDataElement:
>>> assert 'bboxes' not in gt_instances.metainfo_keys()
>>> print(gt_instances.bboxes)
# delete and change property
>>> # delete and change property
>>> gt_instances = BaseDataElement(
metainfo=dict(img_id=0, img_shape=(640, 640))，
data=dict(bboxes=torch.rand((6, 4)), scores=torch.rand((6,))))
... metainfo=dict(img_id=0, img_shape=(640, 640)),
... data=dict(bboxes=torch.rand((6, 4)), scores=torch.rand((6,))))
>>> gt_instances.img_shape = (1280, 1280)
>>> gt_instances.img_shape # (1280, 1280)
>>> gt_instances.bboxes = gt_instances.bboxes * 2
>>> gt_instances.get('img_shape', None) # (640， 640)
>>> gt_instances.get('img_shape', None) # (640, 640)
>>> gt_instances.get('bboxes', None) # 6x4 tensor
>>> del gt_instances.img_shape
>>> del gt_instances.bboxes
@ -113,18 +110,18 @@ class BaseDataElement:
>>> gt_instances.pop('img_shape', None) # None
>>> gt_instances.pop('bboxes', None) # None
# Tensor-like
>>> # Tensor-like
>>> cuda_instances = gt_instances.cuda()
>>> cuda_instances = gt_instancess.to('cuda:0')
>>> cpu_instances = cuda_instances.cpu()
>>> cpu_instances = cuda_instances.to('cpu')
>>> fp16_instances = cuda_instances.to(
device=None, dtype=torch.float16, non_blocking=False, copy=False,
memory_format=torch.preserve_format)
... device=None, dtype=torch.float16, non_blocking=False, copy=False,
... memory_format=torch.preserve_format)
>>> cpu_instances = cuda_instances.detach()
>>> np_instances = cpu_instances.numpy()
# print
>>> # print
>>> img_meta = dict(img_shape=(800, 1196, 3), pad_shape=(800, 1216, 3))
>>> instance_data = BaseDataElement(metainfo=img_meta)
>>> instance_data.det_labels = torch.LongTensor([0, 1, 2, 3])

119
mmengine/data/base_data_sample.py
View File

@ -54,55 +54,52 @@ class BaseDataSample:
Examples:
>>> from mmengine.data import BaseDataElement, BaseDataSample
>>> gt_instances = BaseDataSample()
>>> bboxes = torch.rand((5, 4))
>>> scores = torch.rand((5,))
>>> img_id = 0
>>> img_shape = (800, 1333)
>>> gt_instances = BaseDataElement(
metainfo=dict(img_id=img_id, img_shape=img_shape),
data=dict(bboxes=bboxes, scores=scores))
... metainfo=dict(img_id=img_id, img_shape=img_shape),
... data=dict(bboxes=bboxes, scores=scores))
>>> data = dict(gt_instances=gt_instances)
>>> sample = BaseDataSample(
metainfo=dict(img_id=img_id, img_shape=img_shape),
data=data)
... metainfo=dict(img_id=img_id, img_shape=img_shape),
... data=data)
>>> sample = BaseDataSample(dict(img_id=img_id,
img_shape=(H, W)))
# new
... img_shape=(H, W)))
>>> # new
>>> data1 = dict(bboxes=torch.rand((5, 4)),
scores=torch.rand((5,)))
>>> metainfo1 = dict(img_id=1, img_shape=(640, 640)),
>>> gt_instances1 = BaseDataElement(
metainfo=metainfo1,
data=data1)
... metainfo=metainfo1,
... data=data1)
>>> sample1 = sample.new(
metainfo=metainfo1
data=dict(gt_instances1=gt_instances1)),
... metainfo=metainfo1
... data=dict(gt_instances1=gt_instances1)),
>>> gt_instances2 = gt_instances1.new()
# property add and access
>>> # property add and access
>>> sample = BaseDataSample()
>>> gt_instances = BaseDataElement(
metainfo=dict(img_id=9, img_shape=(100, 100)),
data=dict(bboxes=torch.rand((5, 4)), scores=torch.rand((5,)))
... metainfo=dict(img_id=9, img_shape=(100, 100)),
... data=dict(bboxes=torch.rand((5, 4)), scores=torch.rand((5,)))
>>> sample.set_metainfo(dict(img_id=9, img_shape=(100, 100))
>>> assert 'img_shape' in sample.metainfo_keys()
>>> assert 'img_shape' in sample
>>> assert 'img_shape' not in sample.data_keys()
>>> assert 'img_shape' in sample.keys()
>>> print(sample.img_shape)
>>> gt_instances.gt_instances = gt_instances
>>> assert 'gt_instances' in sample.data_keys()
>>> assert 'gt_instances' in sample
>>> assert 'gt_instances' in sample.keys()
>>> assert 'gt_instances' not in sample.metainfo_keys()
>>> print(sample.gt_instances)
>>> pred_instances = BaseDataElement(
metainfo=dict(img_id=9, img_shape=(100, 100)),
data=dict(bboxes=torch.rand((5, 4)), scores=torch.rand((5,))
... metainfo=dict(img_id=9, img_shape=(100, 100)),
... data=dict(bboxes=torch.rand((5, 4)), scores=torch.rand((5,))
>>> sample.pred_instances = pred_instances
>>> assert 'pred_instances' in sample.data_keys()
>>> assert 'pred_instances' in sample
@ -110,17 +107,17 @@ class BaseDataSample:
>>> assert 'pred_instances' not in sample.metainfo_keys()
>>> print(sample.pred_instances)
# property delete and change
>>> # property delete and change
>>> metainfo=dict(img_id=0, img_shape=(640, 640)
>>> gt_instances = BaseDataElement(
metainfo=metainfo)，
data=dict(bboxes=torch.rand((6, 4)), scores=torch.rand((6,))))
... metainfo=metainfo),
... data=dict(bboxes=torch.rand((6, 4)), scores=torch.rand((6,))))
>>> sample = BaseDataSample(metainfo=metainfo,
data=dict(gt_instances=gt_instances))
... data=dict(gt_instances=gt_instances))
>>> sample.img_shape = (1280, 1280)
>>> sample.img_shape # (1280, 1280)
>>> sample.gt_instances = gt_instances
>>> sample.get('img_shape', None) # (640， 640)
>>> sample.get('img_shape', None) # (640, 640)
>>> sample.get('gt_instances', None)
>>> del sample.img_shape
>>> del sample.gt_instances
@ -129,23 +126,22 @@ class BaseDataSample:
>>> sample.pop('img_shape', None) # None
>>> sample.pop('gt_instances', None) # None
# Tensor-like
>>> # Tensor-like
>>> cuda_sample = gt_instasamplences.cuda()
>>> cuda_sample = gt_sample.to('cuda:0')
>>> cpu_sample = cuda_sample.cpu()
>>> cpu_sample = cuda_sample.to('cpu')
>>> fp16_sample = cuda_sample.to(
device=None, dtype=torch.float16, non_blocking=False, copy=False,
memory_format=torch.preserve_format)
... device=None, dtype=torch.float16, non_blocking=False, copy=False,
... memory_format=torch.preserve_format)
>>> cpu_sample = cuda_sample.detach()
>>> np_sample = cpu_sample.numpy()
# print
>>> # print
>>> metainfo = dict(img_shape=(800, 1196, 3))
>>> gt_instances = BaseDataElement(
metainfo=metainfo,
data=dict(det_labels=torch.LongTensor([0, 1, 2, 3])))
... metainfo=metainfo,
... data=dict(det_labels=torch.LongTensor([0, 1, 2, 3])))
>>> data = dict(gt_instances=gt_instances)
>>> sample = BaseDataSample(metainfo=metainfo, data=data)
>>> print(sample)
@ -161,37 +157,36 @@ class BaseDataSample:
) at 0x7f9705daecd0>'
) at 0x7f981e41c550>'
# inheritance
>>> # inheritance
>>> class DetDataSample(BaseDataSample):
>>> proposals = property(
>>> fget=partial(BaseDataSample.get_field, name='_proposals'),
>>> fset=partial(
>>> BaseDataSample.set_field,
>>> name='_proposals',
>>> dtype=BaseDataElement),
>>> fdel=partial(BaseDataSample.del_field, name='_proposals'),
>>> doc='Region proposals of an image')
>>> gt_instances = property(
>>> fget=partial(BaseDataSample.get_field,
name='_gt_instances'),
>>> fset=partial(
>>> BaseDataSample.set_field,
>>> name='_gt_instances',
>>> dtype=BaseDataElement),
>>> fdel=partial(BaseDataSample.del_field,
name='_gt_instances'),
>>> doc='Ground truth instances of an image')
>>> pred_instances = property(
>>> fget=partial(
>>> BaseDataSample.get_field, name='_pred_instances'),
>>> fset=partial(
>>> BaseDataSample.set_field,
>>> name='_pred_instances',
>>> dtype=BaseDataElement),
>>> fdel=partial(
>>> BaseDataSample.del_field, name='_pred_instances'),
>>> doc='Predicted instances of an image')
... proposals = property(
... fget=partial(BaseDataSample.get_field, name='_proposals'),
... fset=partial(
... BaseDataSample.set_field,
... name='_proposals',
... dtype=BaseDataElement),
... fdel=partial(BaseDataSample.del_field, name='_proposals'),
... doc='Region proposals of an image')
... gt_instances = property(
... fget=partial(BaseDataSample.get_field,
... name='_gt_instances'),
... fset=partial(
... BaseDataSample.set_field,
... name='_gt_instances',
... dtype=BaseDataElement),
... fdel=partial(BaseDataSample.del_field,
... name='_gt_instances'),
... doc='Ground truth instances of an image')
... pred_instances = property(
... fget=partial(
... BaseDataSample.get_field, name='_pred_instances'),
... fset=partial(
... BaseDataSample.set_field,
... name='_pred_instances',
... dtype=BaseDataElement),
... fdel=partial(
... BaseDataSample.del_field, name='_pred_instances'),
... doc='Predicted instances of an image')
>>> det_sample = DetDataSample()
>>> proposals = BaseDataElement(data=dict(bboxes=torch.rand((5, 4))))
>>> det_sample.proposals = proposals
@ -200,7 +195,7 @@ class BaseDataSample:
>>> del det_sample.proposals
>>> assert 'proposals' not in det_sample
>>> with self.assertRaises(AssertionError):
det_sample.proposals = torch.rand((5, 4))
... det_sample.proposals = torch.rand((5, 4))
"""
def __init__(self,

8
tests/test_data/test_data_sample.py
View File

@ -185,6 +185,14 @@ class TestBaseDataSample(TestCase):
instances = BaseDataSample(metainfo, data)
new_metainfo, new_data = self.setup_data()
# avoid generating same metainfo
while True:
if new_metainfo['img_id'] == metainfo['img_id'] or new_metainfo[
'img_shape'] == metainfo['img_shape']:
new_metainfo, new_data = self.setup_data()
else:
break
instances.gt_instances = new_data['gt_instances']
instances.pred_instances = new_data['pred_instances']