596 lines
21 KiB
Python
596 lines
21 KiB
Python
# Copyright (c) OpenMMLab. All rights reserved.
|
|
# import os
|
|
import os.path as osp
|
|
from unittest import TestCase
|
|
from unittest.mock import MagicMock
|
|
|
|
import numpy as np
|
|
# import pickle
|
|
# import tempfile
|
|
from mmengine.registry import TRANSFORMS
|
|
|
|
from mmcls.datasets import DATASETS
|
|
from mmcls.utils import get_root_logger
|
|
|
|
# import torch
|
|
|
|
mmcls_logger = get_root_logger()
|
|
ASSETS_ROOT = osp.abspath(
|
|
osp.join(osp.dirname(__file__), '../../data/dataset'))
|
|
|
|
|
|
class TestBaseDataset(TestCase):
|
|
DATASET_TYPE = 'BaseDataset'
|
|
|
|
DEFAULT_ARGS = dict(data_root=ASSETS_ROOT, ann_file='ann.json')
|
|
|
|
def test_initialize(self):
|
|
dataset_class = DATASETS.get(self.DATASET_TYPE)
|
|
|
|
# Test loading metainfo from ann_file
|
|
cfg = {**self.DEFAULT_ARGS, 'metainfo': None, 'classes': None}
|
|
dataset = dataset_class(**cfg)
|
|
self.assertEqual(
|
|
dataset.CLASSES,
|
|
dataset_class.METAINFO.get('CLASSES', ('first', 'second')))
|
|
self.assertFalse(dataset.test_mode)
|
|
|
|
# Test overriding metainfo by `metainfo` argument
|
|
cfg = {**self.DEFAULT_ARGS, 'metainfo': {'CLASSES': ('bus', 'car')}}
|
|
dataset = dataset_class(**cfg)
|
|
self.assertEqual(dataset.CLASSES, ('bus', 'car'))
|
|
|
|
# Test overriding metainfo by `classes` argument
|
|
cfg = {**self.DEFAULT_ARGS, 'classes': ['bus', 'car']}
|
|
dataset = dataset_class(**cfg)
|
|
self.assertEqual(dataset.CLASSES, ('bus', 'car'))
|
|
|
|
classes_file = osp.join(ASSETS_ROOT, 'classes.txt')
|
|
cfg = {**self.DEFAULT_ARGS, 'classes': classes_file}
|
|
dataset = dataset_class(**cfg)
|
|
self.assertEqual(dataset.CLASSES, ('bus', 'car'))
|
|
self.assertEqual(dataset.class_to_idx, {'bus': 0, 'car': 1})
|
|
|
|
# Test invalid classes
|
|
cfg = {**self.DEFAULT_ARGS, 'classes': dict(classes=1)}
|
|
with self.assertRaisesRegex(ValueError, "type <class 'dict'>"):
|
|
dataset_class(**cfg)
|
|
|
|
def test_get_cat_ids(self):
|
|
dataset_class = DATASETS.get(self.DATASET_TYPE)
|
|
dataset = dataset_class(**self.DEFAULT_ARGS)
|
|
|
|
cat_ids = dataset.get_cat_ids(0)
|
|
self.assertIsInstance(cat_ids, list)
|
|
self.assertEqual(len(cat_ids), 1)
|
|
self.assertIsInstance(cat_ids[0], int)
|
|
|
|
def test_repr(self):
|
|
dataset_class = DATASETS.get(self.DATASET_TYPE)
|
|
cfg = {**self.DEFAULT_ARGS, 'lazy_init': True}
|
|
dataset = dataset_class(**cfg)
|
|
|
|
head = 'Dataset ' + dataset.__class__.__name__
|
|
self.assertIn(head, repr(dataset))
|
|
|
|
if dataset.CLASSES is not None:
|
|
num_classes = len(dataset.CLASSES)
|
|
self.assertIn(f'Number of categories: \t{num_classes}',
|
|
repr(dataset))
|
|
else:
|
|
self.assertIn('The `CLASSES` meta info is not set.', repr(dataset))
|
|
|
|
self.assertIn("Haven't been initialized", repr(dataset))
|
|
dataset.full_init()
|
|
self.assertIn(f'Number of samples: \t{len(dataset)}', repr(dataset))
|
|
|
|
self.assertIn(f'Annotation file: \t{dataset.ann_file}', repr(dataset))
|
|
self.assertIn(f'Prefix of images: \t{dataset.img_prefix}',
|
|
repr(dataset))
|
|
|
|
TRANSFORMS.register_module(name='test_mock', module=MagicMock)
|
|
cfg = {**self.DEFAULT_ARGS, 'pipeline': [dict(type='test_mock')]}
|
|
dataset = dataset_class(**cfg)
|
|
self.assertIn('With transforms', repr(dataset))
|
|
del TRANSFORMS.module_dict['test_mock']
|
|
|
|
|
|
"""Temporarily disabled.
|
|
class TestMultiLabelDataset(TestBaseDataset):
|
|
DATASET_TYPE = 'MultiLabelDataset'
|
|
|
|
def test_get_cat_ids(self):
|
|
dataset_class = DATASETS.get(self.DATASET_TYPE)
|
|
fake_ann = [
|
|
dict(
|
|
img_prefix='',
|
|
img_info=dict(),
|
|
gt_label=np.array([0, 1, 1, 0], dtype=np.uint8))
|
|
]
|
|
|
|
with patch.object(dataset_class, 'load_annotations') as mock_load:
|
|
mock_load.return_value = fake_ann
|
|
dataset = dataset_class(**self.DEFAULT_ARGS)
|
|
|
|
cat_ids = dataset.get_cat_ids(0)
|
|
self.assertIsInstance(cat_ids, list)
|
|
self.assertEqual(len(cat_ids), 2)
|
|
self.assertIsInstance(cat_ids[0], int)
|
|
self.assertEqual(cat_ids, [1, 2])
|
|
|
|
def test_evaluate(self):
|
|
dataset_class = DATASETS.get(self.DATASET_TYPE)
|
|
|
|
fake_ann = [
|
|
dict(gt_label=np.array([1, 1, 0, -1], dtype=np.int8)),
|
|
dict(gt_label=np.array([1, 1, 0, -1], dtype=np.int8)),
|
|
dict(gt_label=np.array([0, -1, 1, -1], dtype=np.int8)),
|
|
dict(gt_label=np.array([0, 1, 0, -1], dtype=np.int8)),
|
|
dict(gt_label=np.array([0, 1, 0, -1], dtype=np.int8)),
|
|
]
|
|
|
|
with patch.object(dataset_class, 'load_annotations') as mock_load:
|
|
mock_load.return_value = fake_ann
|
|
dataset = dataset_class(**self.DEFAULT_ARGS)
|
|
|
|
fake_results = np.array([
|
|
[0.9, 0.8, 0.3, 0.2],
|
|
[0.1, 0.2, 0.2, 0.1],
|
|
[0.7, 0.5, 0.9, 0.3],
|
|
[0.8, 0.1, 0.1, 0.2],
|
|
[0.8, 0.1, 0.1, 0.2],
|
|
])
|
|
|
|
# the metric must be valid for the dataset
|
|
with self.assertRaisesRegex(ValueError,
|
|
"{'unknown'} is not supported"):
|
|
dataset.evaluate(fake_results, metric='unknown')
|
|
|
|
# only one metric
|
|
eval_results = dataset.evaluate(fake_results, metric='mAP')
|
|
self.assertEqual(eval_results.keys(), {'mAP'})
|
|
self.assertAlmostEqual(eval_results['mAP'], 67.5, places=4)
|
|
|
|
# multiple metrics
|
|
eval_results = dataset.evaluate(
|
|
fake_results, metric=['mAP', 'CR', 'OF1'])
|
|
self.assertEqual(eval_results.keys(), {'mAP', 'CR', 'OF1'})
|
|
self.assertAlmostEqual(eval_results['mAP'], 67.50, places=2)
|
|
self.assertAlmostEqual(eval_results['CR'], 43.75, places=2)
|
|
self.assertAlmostEqual(eval_results['OF1'], 42.86, places=2)
|
|
"""
|
|
|
|
|
|
class TestCustomDataset(TestBaseDataset):
|
|
DATASET_TYPE = 'CustomDataset'
|
|
|
|
DEFAULT_ARGS = dict(data_root=ASSETS_ROOT, ann_file='ann.txt')
|
|
|
|
def test_initialize(self):
|
|
dataset_class = DATASETS.get(self.DATASET_TYPE)
|
|
|
|
# Test overriding metainfo by `metainfo` argument
|
|
cfg = {**self.DEFAULT_ARGS, 'metainfo': {'CLASSES': ('bus', 'car')}}
|
|
dataset = dataset_class(**cfg)
|
|
self.assertEqual(dataset.CLASSES, ('bus', 'car'))
|
|
|
|
# Test overriding metainfo by `classes` argument
|
|
cfg = {**self.DEFAULT_ARGS, 'classes': ['bus', 'car']}
|
|
dataset = dataset_class(**cfg)
|
|
self.assertEqual(dataset.CLASSES, ('bus', 'car'))
|
|
|
|
classes_file = osp.join(ASSETS_ROOT, 'classes.txt')
|
|
cfg = {**self.DEFAULT_ARGS, 'classes': classes_file}
|
|
dataset = dataset_class(**cfg)
|
|
self.assertEqual(dataset.CLASSES, ('bus', 'car'))
|
|
self.assertEqual(dataset.class_to_idx, {'bus': 0, 'car': 1})
|
|
|
|
# Test invalid classes
|
|
cfg = {**self.DEFAULT_ARGS, 'classes': dict(classes=1)}
|
|
with self.assertRaisesRegex(ValueError, "type <class 'dict'>"):
|
|
dataset_class(**cfg)
|
|
|
|
def test_load_data_list(self):
|
|
dataset_class = DATASETS.get(self.DATASET_TYPE)
|
|
|
|
# test load without ann_file
|
|
cfg = {
|
|
**self.DEFAULT_ARGS,
|
|
'data_prefix': ASSETS_ROOT,
|
|
'ann_file': None,
|
|
}
|
|
dataset = dataset_class(**cfg)
|
|
self.assertEqual(len(dataset), 3)
|
|
self.assertEqual(dataset.CLASSES, ('a', 'b')) # auto infer classes
|
|
self.assertGreaterEqual(
|
|
dataset.get_data_info(0).items(), {
|
|
'img_path': osp.join(ASSETS_ROOT, 'a/1.JPG'),
|
|
'gt_label': 0
|
|
}.items())
|
|
self.assertGreaterEqual(
|
|
dataset.get_data_info(2).items(), {
|
|
'img_path': osp.join(ASSETS_ROOT, 'b/subb/3.jpg'),
|
|
'gt_label': 1
|
|
}.items())
|
|
|
|
# test ann_file assertion
|
|
cfg = {
|
|
**self.DEFAULT_ARGS,
|
|
'data_prefix': ASSETS_ROOT,
|
|
'ann_file': ['ann_file.txt'],
|
|
}
|
|
with self.assertRaisesRegex(TypeError, 'expected str'):
|
|
dataset_class(**cfg)
|
|
|
|
# test load with ann_file
|
|
cfg = {
|
|
**self.DEFAULT_ARGS,
|
|
'data_root': ASSETS_ROOT,
|
|
'ann_file': 'ann.txt',
|
|
}
|
|
dataset = dataset_class(**cfg)
|
|
self.assertEqual(len(dataset), 3)
|
|
# custom dataset won't infer CLASSES from ann_file
|
|
self.assertIsNone(dataset.CLASSES, None)
|
|
self.assertGreaterEqual(
|
|
dataset.get_data_info(0).items(), {
|
|
'img_path': osp.join(ASSETS_ROOT, 'a/1.JPG'),
|
|
'gt_label': 0,
|
|
}.items())
|
|
self.assertGreaterEqual(
|
|
dataset.get_data_info(2).items(), {
|
|
'img_path': osp.join(ASSETS_ROOT, 'b/subb/3.jpg'),
|
|
'gt_label': 1
|
|
}.items())
|
|
np.testing.assert_equal(dataset.get_gt_labels(), np.array([0, 1, 1]))
|
|
|
|
# test load with absolute ann_file
|
|
cfg = {
|
|
**self.DEFAULT_ARGS,
|
|
'data_root': None,
|
|
'data_prefix': None,
|
|
'ann_file': osp.join(ASSETS_ROOT, 'ann.txt'),
|
|
}
|
|
dataset = dataset_class(**cfg)
|
|
self.assertEqual(len(dataset), 3)
|
|
# custom dataset won't infer CLASSES from ann_file
|
|
self.assertIsNone(dataset.CLASSES, None)
|
|
self.assertGreaterEqual(
|
|
dataset.get_data_info(0).items(), {
|
|
'img_path': 'a/1.JPG',
|
|
'gt_label': 0,
|
|
}.items())
|
|
self.assertGreaterEqual(
|
|
dataset.get_data_info(2).items(), {
|
|
'img_path': 'b/subb/3.jpg',
|
|
'gt_label': 1
|
|
}.items())
|
|
|
|
# test extensions filter
|
|
cfg = {
|
|
**self.DEFAULT_ARGS, 'data_prefix': dict(img_path=ASSETS_ROOT),
|
|
'ann_file': None,
|
|
'extensions': ('.txt', )
|
|
}
|
|
with self.assertRaisesRegex(RuntimeError,
|
|
'Supported extensions are: .txt'):
|
|
dataset_class(**cfg)
|
|
|
|
cfg = {
|
|
**self.DEFAULT_ARGS, 'data_prefix': ASSETS_ROOT,
|
|
'ann_file': None,
|
|
'extensions': ('.jpeg', )
|
|
}
|
|
with self.assertLogs(mmcls_logger, 'WARN') as log:
|
|
dataset = dataset_class(**cfg)
|
|
self.assertIn('Supported extensions are: .jpeg', log.output[0])
|
|
self.assertEqual(len(dataset), 1)
|
|
self.assertGreaterEqual(
|
|
dataset.get_data_info(0).items(), {
|
|
'img_path': osp.join(ASSETS_ROOT, 'b/2.jpeg'),
|
|
'gt_label': 1
|
|
}.items())
|
|
|
|
# test classes check
|
|
cfg = {
|
|
**self.DEFAULT_ARGS,
|
|
'data_prefix': ASSETS_ROOT,
|
|
'classes': ('apple', 'banana'),
|
|
'ann_file': None,
|
|
}
|
|
dataset = dataset_class(**cfg)
|
|
self.assertEqual(dataset.CLASSES, ('apple', 'banana'))
|
|
|
|
cfg['classes'] = ['apple', 'banana', 'dog']
|
|
with self.assertRaisesRegex(AssertionError,
|
|
r"\(2\) doesn't match .* classes \(3\)"):
|
|
dataset_class(**cfg)
|
|
|
|
|
|
class TestImageNet(TestCustomDataset):
|
|
DATASET_TYPE = 'ImageNet'
|
|
|
|
DEFAULT_ARGS = dict(data_root=ASSETS_ROOT, ann_file='ann.txt')
|
|
|
|
def test_load_data_list(self):
|
|
dataset_class = DATASETS.get(self.DATASET_TYPE)
|
|
|
|
# test classes number
|
|
cfg = {
|
|
**self.DEFAULT_ARGS,
|
|
'data_prefix': ASSETS_ROOT,
|
|
'ann_file': None,
|
|
}
|
|
with self.assertRaisesRegex(
|
|
AssertionError, r"\(2\) doesn't match .* classes \(1000\)"):
|
|
dataset_class(**cfg)
|
|
|
|
# test override classes
|
|
cfg = {
|
|
**self.DEFAULT_ARGS,
|
|
'data_prefix': ASSETS_ROOT,
|
|
'classes': ['cat', 'dog'],
|
|
'ann_file': None,
|
|
}
|
|
dataset = dataset_class(**cfg)
|
|
self.assertEqual(len(dataset), 3)
|
|
self.assertEqual(dataset.CLASSES, ('cat', 'dog'))
|
|
|
|
|
|
class TestImageNet21k(TestCustomDataset):
|
|
DATASET_TYPE = 'ImageNet21k'
|
|
|
|
DEFAULT_ARGS = dict(
|
|
data_root=ASSETS_ROOT, classes=['cat', 'dog'], ann_file='ann.txt')
|
|
|
|
def test_load_data_list(self):
|
|
super().test_initialize()
|
|
dataset_class = DATASETS.get(self.DATASET_TYPE)
|
|
|
|
# The multi_label option is not implemented not.
|
|
cfg = {**self.DEFAULT_ARGS, 'multi_label': True}
|
|
with self.assertRaisesRegex(NotImplementedError, 'not supported'):
|
|
dataset_class(**cfg)
|
|
|
|
# Warn about ann_file
|
|
cfg = {**self.DEFAULT_ARGS, 'ann_file': None}
|
|
with self.assertLogs(mmcls_logger, 'WARN') as log:
|
|
dataset_class(**cfg)
|
|
self.assertIn('specify the `ann_file`', log.output[0])
|
|
|
|
# Warn about classes
|
|
cfg = {**self.DEFAULT_ARGS, 'classes': None}
|
|
with self.assertLogs(mmcls_logger, 'WARN') as log:
|
|
dataset_class(**cfg)
|
|
self.assertIn('specify the `classes`', log.output[0])
|
|
|
|
|
|
"""Temporarily disabled.
|
|
|
|
class TestMNIST(TestBaseDataset):
|
|
DATASET_TYPE = 'MNIST'
|
|
|
|
@classmethod
|
|
def setUpClass(cls) -> None:
|
|
super().setUpClass()
|
|
|
|
tmpdir = tempfile.TemporaryDirectory()
|
|
cls.tmpdir = tmpdir
|
|
data_prefix = tmpdir.name
|
|
cls.DEFAULT_ARGS = dict(data_prefix=data_prefix, pipeline=[])
|
|
|
|
dataset_class = DATASETS.get(cls.DATASET_TYPE)
|
|
|
|
def rm_suffix(s):
|
|
return s[:s.rfind('.')]
|
|
|
|
train_image_file = osp.join(
|
|
data_prefix,
|
|
rm_suffix(dataset_class.resources['train_image_file'][0]))
|
|
train_label_file = osp.join(
|
|
data_prefix,
|
|
rm_suffix(dataset_class.resources['train_label_file'][0]))
|
|
test_image_file = osp.join(
|
|
data_prefix,
|
|
rm_suffix(dataset_class.resources['test_image_file'][0]))
|
|
test_label_file = osp.join(
|
|
data_prefix,
|
|
rm_suffix(dataset_class.resources['test_label_file'][0]))
|
|
cls.fake_img = np.random.randint(0, 255, size=(28, 28), dtype=np.uint8)
|
|
cls.fake_label = np.random.randint(0, 10, size=(1, ), dtype=np.uint8)
|
|
|
|
for file in [train_image_file, test_image_file]:
|
|
magic = b'\x00\x00\x08\x03' # num_dims = 3, type = uint8
|
|
head = b'\x00\x00\x00\x01' + b'\x00\x00\x00\x1c' * 2 # (1, 28, 28)
|
|
data = magic + head + cls.fake_img.flatten().tobytes()
|
|
with open(file, 'wb') as f:
|
|
f.write(data)
|
|
|
|
for file in [train_label_file, test_label_file]:
|
|
magic = b'\x00\x00\x08\x01' # num_dims = 3, type = uint8
|
|
head = b'\x00\x00\x00\x01' # (1, )
|
|
data = magic + head + cls.fake_label.tobytes()
|
|
with open(file, 'wb') as f:
|
|
f.write(data)
|
|
|
|
def test_load_annotations(self):
|
|
dataset_class = DATASETS.get(self.DATASET_TYPE)
|
|
|
|
with patch.object(dataset_class, 'download'):
|
|
# Test default behavior
|
|
dataset = dataset_class(**self.DEFAULT_ARGS)
|
|
self.assertEqual(len(dataset), 1)
|
|
|
|
data_info = dataset[0]
|
|
np.testing.assert_equal(data_info['img'], self.fake_img)
|
|
np.testing.assert_equal(data_info['gt_label'], self.fake_label)
|
|
|
|
@classmethod
|
|
def tearDownClass(cls):
|
|
cls.tmpdir.cleanup()
|
|
|
|
class TestCIFAR10(TestBaseDataset):
|
|
DATASET_TYPE = 'CIFAR10'
|
|
|
|
@classmethod
|
|
def setUpClass(cls) -> None:
|
|
super().setUpClass()
|
|
|
|
tmpdir = tempfile.TemporaryDirectory()
|
|
cls.tmpdir = tmpdir
|
|
data_prefix = tmpdir.name
|
|
cls.DEFAULT_ARGS = dict(data_prefix=data_prefix, pipeline=[])
|
|
|
|
dataset_class = DATASETS.get(cls.DATASET_TYPE)
|
|
base_folder = osp.join(data_prefix, dataset_class.base_folder)
|
|
os.mkdir(base_folder)
|
|
|
|
cls.fake_imgs = np.random.randint(
|
|
0, 255, size=(6, 3 * 32 * 32), dtype=np.uint8)
|
|
cls.fake_labels = np.random.randint(0, 10, size=(6, ))
|
|
cls.fake_classes = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
|
|
|
|
batch1 = dict(
|
|
data=cls.fake_imgs[:2], labels=cls.fake_labels[:2].tolist())
|
|
with open(osp.join(base_folder, 'data_batch_1'), 'wb') as f:
|
|
f.write(pickle.dumps(batch1))
|
|
|
|
batch2 = dict(
|
|
data=cls.fake_imgs[2:4], labels=cls.fake_labels[2:4].tolist())
|
|
with open(osp.join(base_folder, 'data_batch_2'), 'wb') as f:
|
|
f.write(pickle.dumps(batch2))
|
|
|
|
test_batch = dict(
|
|
data=cls.fake_imgs[4:], labels=cls.fake_labels[4:].tolist())
|
|
with open(osp.join(base_folder, 'test_batch'), 'wb') as f:
|
|
f.write(pickle.dumps(test_batch))
|
|
|
|
meta = {dataset_class.meta['key']: cls.fake_classes}
|
|
meta_filename = dataset_class.meta['filename']
|
|
with open(osp.join(base_folder, meta_filename), 'wb') as f:
|
|
f.write(pickle.dumps(meta))
|
|
|
|
dataset_class.train_list = [['data_batch_1', None],
|
|
['data_batch_2', None]]
|
|
dataset_class.test_list = [['test_batch', None]]
|
|
dataset_class.meta['md5'] = None
|
|
|
|
def test_load_annotations(self):
|
|
dataset_class = DATASETS.get(self.DATASET_TYPE)
|
|
|
|
# Test default behavior
|
|
dataset = dataset_class(**self.DEFAULT_ARGS)
|
|
self.assertEqual(len(dataset), 4)
|
|
self.assertEqual(dataset.CLASSES, self.fake_classes)
|
|
|
|
data_info = dataset[0]
|
|
fake_img = self.fake_imgs[0].reshape(3, 32, 32).transpose(1, 2, 0)
|
|
np.testing.assert_equal(data_info['img'], fake_img)
|
|
np.testing.assert_equal(data_info['gt_label'], self.fake_labels[0])
|
|
|
|
# Test with test_mode=True
|
|
cfg = {**self.DEFAULT_ARGS, 'test_mode': True}
|
|
dataset = dataset_class(**cfg)
|
|
self.assertEqual(len(dataset), 2)
|
|
|
|
data_info = dataset[0]
|
|
fake_img = self.fake_imgs[4].reshape(3, 32, 32).transpose(1, 2, 0)
|
|
np.testing.assert_equal(data_info['img'], fake_img)
|
|
np.testing.assert_equal(data_info['gt_label'], self.fake_labels[4])
|
|
|
|
@classmethod
|
|
def tearDownClass(cls):
|
|
cls.tmpdir.cleanup()
|
|
|
|
class TestCIFAR100(TestCIFAR10):
|
|
DATASET_TYPE = 'CIFAR100'
|
|
|
|
class TestVOC(TestMultiLabelDataset):
|
|
DATASET_TYPE = 'VOC'
|
|
|
|
DEFAULT_ARGS = dict(data_prefix='VOC2007', pipeline=[])
|
|
|
|
class TestCUB(TestBaseDataset):
|
|
DATASET_TYPE = 'CUB'
|
|
|
|
@classmethod
|
|
def setUpClass(cls) -> None:
|
|
super().setUpClass()
|
|
|
|
tmpdir = tempfile.TemporaryDirectory()
|
|
cls.tmpdir = tmpdir
|
|
cls.data_prefix = tmpdir.name
|
|
cls.ann_file = osp.join(cls.data_prefix, 'ann_file.txt')
|
|
cls.image_class_labels_file = osp.join(cls.data_prefix, 'classes.txt')
|
|
cls.train_test_split_file = osp.join(cls.data_prefix, 'split.txt')
|
|
cls.train_test_split_file2 = osp.join(cls.data_prefix, 'split2.txt')
|
|
cls.DEFAULT_ARGS = dict(
|
|
data_prefix=cls.data_prefix,
|
|
pipeline=[],
|
|
ann_file=cls.ann_file,
|
|
image_class_labels_file=cls.image_class_labels_file,
|
|
train_test_split_file=cls.train_test_split_file)
|
|
|
|
with open(cls.ann_file, 'w') as f:
|
|
f.write('\n'.join([
|
|
'1 1.txt',
|
|
'2 2.txt',
|
|
'3 3.txt',
|
|
]))
|
|
|
|
with open(cls.image_class_labels_file, 'w') as f:
|
|
f.write('\n'.join([
|
|
'1 2',
|
|
'2 3',
|
|
'3 1',
|
|
]))
|
|
|
|
with open(cls.train_test_split_file, 'w') as f:
|
|
f.write('\n'.join([
|
|
'1 0',
|
|
'2 1',
|
|
'3 1',
|
|
]))
|
|
|
|
with open(cls.train_test_split_file2, 'w') as f:
|
|
f.write('\n'.join([
|
|
'1 0',
|
|
'2 1',
|
|
]))
|
|
|
|
def test_load_annotations(self):
|
|
dataset_class = DATASETS.get(self.DATASET_TYPE)
|
|
|
|
# Test default behavior
|
|
dataset = dataset_class(**self.DEFAULT_ARGS)
|
|
self.assertEqual(len(dataset), 2)
|
|
self.assertEqual(dataset.CLASSES, dataset_class.CLASSES)
|
|
|
|
data_info = dataset[0]
|
|
np.testing.assert_equal(data_info['img_prefix'], self.data_prefix)
|
|
np.testing.assert_equal(data_info['img_info'], {'filename': '2.txt'})
|
|
np.testing.assert_equal(data_info['gt_label'], 3 - 1)
|
|
|
|
# Test with test_mode=True
|
|
cfg = {**self.DEFAULT_ARGS, 'test_mode': True}
|
|
dataset = dataset_class(**cfg)
|
|
self.assertEqual(len(dataset), 1)
|
|
|
|
data_info = dataset[0]
|
|
np.testing.assert_equal(data_info['img_prefix'], self.data_prefix)
|
|
np.testing.assert_equal(data_info['img_info'], {'filename': '1.txt'})
|
|
np.testing.assert_equal(data_info['gt_label'], 2 - 1)
|
|
|
|
# Test if the numbers of line are not match
|
|
cfg = {
|
|
**self.DEFAULT_ARGS, 'train_test_split_file':
|
|
self.train_test_split_file2
|
|
}
|
|
with self.assertRaisesRegex(AssertionError, 'should have same length'):
|
|
dataset_class(**cfg)
|
|
|
|
@classmethod
|
|
def tearDownClass(cls):
|
|
cls.tmpdir.cleanup()
|
|
"""
|