def test_center_net_estimator():
from gluoncv.auto.tasks import ObjectDetection
task = ObjectDetection({'num_trials': 1})
detector = task.fit(OBJECT_DETCTION_DATASET)
assert task.fit_summary.get('valid_map', 0) > 0
_, _, test_data = OBJECT_DETCTION_DATASET.random_split()
test_result = detector.predict(test_data)
def _validate_data(self, data):
"""Check whether data is valid, try to convert with best effort if not"""
if len(data) < 1:
raise ValueError('Empty dataset.')
if not (hasattr(data, 'classes') and hasattr(data, 'to_mxnet')):
if isinstance(data, pd.DataFrame):
# raw dataframe, try to add metadata automatically
infer_classes = []
if 'image' in data.columns:
# check image relative/abs path is valid
sample = data.iloc[0]['image']
if not os.path.isfile(sample):
raise OSError(
f'Detected invalid image path `{sample}`, please ensure all image paths are absolute or you are using the right working directory.'
)
if 'rois' in data.columns and 'image' in data.columns:
sample = data.iloc[0]['rois']
for sample_key in ('class', 'xmin', 'ymin', 'xmax',
'ymax'):
assert sample_key in sample, f'key `{sample_key}` required in `rois`'
class_column = data.rois.apply(
lambda x: x.get('class', 'unknown'))
infer_classes = class_column.unique().tolist()
data['rois'] = data['rois'].apply(lambda x: x.update(
{'difficult': x.get('difficult', 0)} or x))
data = _ObjectDetection.Dataset(
data.sort_values('image').reset_index(drop=True),
classes=infer_classes)
elif 'image' in data and 'class' in data and 'xmin' in data and 'ymin' in data and 'xmax' in data and 'ymax' in data:
infer_classes = data['class'].unique().tolist()
if 'difficult' not in data.columns:
data['difficult'] = 0
data = _ObjectDetection.Dataset(
data.sort_values('image').reset_index(drop=True),
classes=infer_classes)
data = data.pack()
data.classes = infer_classes
else:
err_msg = 'Unable to convert raw DataFrame to ObjectDetector Dataset, ' + \
'`image` and `rois` columns are required.' + \
'You may visit `https://auto.gluon.ai/stable/tutorials/object_detection/dataset.html` ' + \
'for details.'
raise AttributeError(err_msg)
logger.log(
20,
'Converting raw DataFrame to ObjectDetector.Dataset...')
logger.log(
20,
f'Detected {len(infer_classes)} unique classes: {infer_classes}'
)
instruction = 'train_data = ObjectDetector.Dataset(train_data, classes=["foo", "bar"])'
logger.log(
20,
f'If you feel the `classes` is inaccurate, please construct the dataset explicitly, e.g. {instruction}'
)
return data
def test_time_out_detection():
time_limit = 30
from gluoncv.auto.tasks import ObjectDetection
task = ObjectDetection({
'num_trials': 1,
'epochs': 50,
'time_limits': time_limit
})
tic = time.time()
detector = task.fit(OBJECT_DETECTION_TRAIN)
# check time_limit with a little bit overhead
assert (time.time() - tic) < time_limit + 180
def test_object_detection_estimator_transfer():
from gluoncv.auto.tasks import ObjectDetection
task = ObjectDetection({
'num_trials':
1,
'transfer':
ag.Categorical('yolo3_darknet53_coco', 'ssd_512_resnet50_v1_voc'),
'estimator':
'ssd'
})
detector = task.fit(OBJECT_DETCTION_DATASET)
assert task.fit_summary().get('valid_map', 0) > 0
_, _, test_data = OBJECT_DETCTION_DATASET.random_split()
test_result = detector.predict(test_data)
def test_object_detection_estimator_transfer():
from gluoncv.auto.tasks import ObjectDetection
task = ObjectDetection({
'num_trials':
1,
'epochs':
1,
'transfer':
ag.Categorical('yolo3_darknet53_coco', 'ssd_512_resnet50_v1_voc'),
'estimator':
'ssd',
'batch_size':
4
})
detector = task.fit(OBJECT_DETECTION_TRAIN)
assert task.fit_summary().get('valid_map', 0) > 0
test_result = detector.predict(OBJECT_DETECTION_TEST)
'batch_size':
ag.Int(3, 6), # [8, 16, 32, 64]
'momentum':
ag.Real(0.85, 0.95),
'wd':
ag.Real(1e-6, 1e-2, log=True),
'epochs':
20,
'num_trials':
args.num_trials,
'search_strategy':
'bayesopt'
}
# specify learning task
task = ObjectDetection(config)
# specify dataset
dataset = Dataset.get(args.dataset)
train_data, valid_data = dataset.split(0.8)
# fit auto estimator
detector = task.fit(train_data, valid_data)
# evaluate auto estimator
eval_map = detector.evaluate(valid_data)
logging.info('evaluation: mAP={}'.format(eval_map[-1][-1]))
# save and load auto estimator
detector.save('ssd_detector.pkl')
detector = ObjectDetection.load('ssd_detector.pkl')
def test_object_detection_estimator():
from gluoncv.auto.tasks import ObjectDetection
task = ObjectDetection({'num_trials': 1, 'epochs': 1})
detector = task.fit(OBJECT_DETECTION_TRAIN)
assert task.fit_summary().get('valid_map', 0) > 0
test_result = detector.predict(OBJECT_DETECTION_TEST)
def test_center_net_estimator():
from gluoncv.auto.tasks import ObjectDetection
task = ObjectDetection({'num_trials': 1})
detector = task.fit(OBJECT_DETCTION_DATASET)
assert task.fit_summary.get('valid_map', 0) > 0
'dataset': 'voc_tiny',
'meta_arch': 'ssd',
'lr': ag.Categorical(1e-3, 5e-4),
'epochs': 5,
'num_trials': 2
}
if 'meta_arch' in args:
if args['meta_arch'] == 'ssd':
estimator = SSDEstimator
elif args['meta_arch'] == 'faster_rcnn':
estimator = FasterRCNNEstimator
elif args['meta_arch'] == 'yolo3':
estimator = YOLOv3Estimator
elif args['meta_arch'] == 'center_net':
estimator = CenterNetEstimator
else:
raise NotImplementedError('%s is not implemented.',
args['meta_arch'])
else:
estimator = None
task = ObjectDetection(args, estimator)
detector = task.fit()
test_map = detector.evaluate()
# print("mAP on test dataset: {}".format(test_map[-1][-1]))
# print(test_map)
detector.save('detector.pkl')
detector = ObjectDetection.load('detector.pkl')