def get_bisenet(dataset='pascal_paper',
backbone='resnet18',
pretrained_base=True,
pretrained=False,
root=os.path.expanduser('~/.torch/models'),
**kwargs):
acronyms = {
'pascal_voc': 'voc',
'pascal_paper': 'voc',
'pascal_aug': 'voc',
'ade20k': 'ade',
'coco': 'coco',
'citys': 'citys',
}
from data import datasets
config = bisenet_spec[backbone]
if config[2] is not None:
feat = nn.ModuleList(
_parse_network(config[1],
outputs=config[2],
pretrained=pretrained_base))
else:
feat = config[1]
model = BiseNet(config[0],
datasets[dataset].NUM_CLASS,
backbone=feat,
**kwargs)
if pretrained:
from model.model_store import get_model_file
model.load_state_dict(
torch.load(
get_model_file('bisenet_%s_%s' % (backbone, acronyms[dataset]),
root=root)))
return model
def centernet_dla34_dcn_coco(pretrained=False,
pretrained_base=False,
**kwargs):
from data.mscoco.detection_cv import COCODetection
classes = COCODetection.CLASSES
pretrained_base = False if pretrained else pretrained_base
name = 'centernet_dla34_dcn_coco'
norm_layer = kwargs.get(
'norm_layer') if 'norm_layer' in kwargs else nn.BatchNorm2d
norm_kwargs = kwargs.get(
'norm_kwargs') if 'norm_kwargs' in kwargs else None
outputs = [[1], [2], [3], [4], [5], [6]]
features = nn.ModuleList(
_parse_network('dla34',
outputs,
pretrained_base,
norm_layer=norm_layer,
norm_kwargs=norm_kwargs))
deconv_layers = DLADeConvLayer([16, 32, 64, 128, 256, 512],
down_ratio=4,
last_level=5,
norm_layer=norm_layer,
norm_kwargs=norm_kwargs)
return get_centernet(name,
features,
deconv_layers,
heads={
'hm': 80,
'wh': 2,
'reg': 2
},
head_conv=256,
classes=classes,
pretrained=pretrained,
**kwargs)
def __init__(self,
nclass,
aux,
backbone='resnet50',
jpu=False,
dilated=True,
height=None,
width=None,
base_size=520,
crop_size=480,
keep_shape=False,
pretrained_base=True,
**kwargs):
super(SegBaseModel, self).__init__()
self.aux = aux
self.nclass = nclass
self.jpu = jpu
self.keep_shape = keep_shape
if isinstance(backbone, torch.nn.ModuleList) and len(backbone) == 3:
self.base1, self.base2, self.base3 = backbone[0], backbone[
1], backbone[2]
else:
if backbone == 'resnet50':
outputs = [[11, 3], [12, 5], [13, 2]]
elif backbone == 'resnet101':
outputs = [[11, 3], [12, 22], [13, 2]]
elif backbone == 'resnet152':
outputs = [[11, 7], [12, 35], [13, 2]]
else:
raise RuntimeError('unknown backbone: {}'.format(backbone))
# TODO: change
self.base1, self.base2, self.base3 = _parse_network(
backbone + '_v1s',
outputs,
pretrained=pretrained_base,
dilated=dilated)
height = height if height is not None else crop_size
width = width if width is not None else crop_size
self._up_kwargs = (height, width)
self.base_size = base_size
self.crop_size = crop_size
if jpu:
self.JPU = JPU([512, 1024, 2048], width=512)
def mask_rcnn_resnet50_v1b_coco(pretrained=False, pretrained_base=True, **kwargs):
r"""Mask RCNN model from the paper
"He, K., Gkioxari, G., Doll&ar, P., & Girshick, R. (2017). Mask R-CNN"
Parameters
----------
pretrained : bool or str
Boolean value controls whether to load the default pretrained weights for model.
String value represents the hashtag for a certain version of pretrained weights.
pretrained_base : bool or str, optional, default is True
Load pretrained base network, the extra layers are randomized. Note that
if pretrained is `True`, this has no effect.
ctx : Context, default CPU
The context in which to load the pretrained weights.
root : str, default '~/.mxnet/models'
Location for keeping the model parameters.
Examples
--------
>>> model = mask_rcnn_resnet50_v1b_coco(pretrained=True)
>>> print(model)
"""
from data.mscoco.detection_cv import COCODetection
classes = COCODetection.CLASSES
pretrained_base = False if pretrained else pretrained_base
outputs = [[6, 5], [7, 2]]
features, top_features = _parse_network('resnet50_v1b', outputs, pretrained_base)
train_patterns = '|'.join(['.*dense', '.*rpn', '.*mask',
'.*down(2|3|4)_conv', '.*layers(2|3|4)_conv'])
return get_mask_rcnn(
name='resnet50_v1b', dataset='coco', pretrained=pretrained,
features=features, top_features=top_features, classes=classes,
mask_channels=256, rcnn_max_dets=1000,
short=800, max_size=1333, train_patterns=train_patterns,
nms_thresh=0.5, nms_topk=-1, post_nms=-1, in_channels=2048,
roi_mode='align', roi_size=(14, 14), strides=16, clip=4.42,
rpn_in_channel=1024, rpn_channel=1024, base_size=16, scales=(2, 4, 8, 16, 32),
ratios=(0.5, 1, 2), alloc_size=(128, 128), rpn_nms_thresh=0.7,
rpn_train_pre_nms=12000, rpn_train_post_nms=2000,
rpn_test_pre_nms=6000, rpn_test_post_nms=1000, rpn_min_size=0,
num_sample=128, pos_iou_thresh=0.5, pos_ratio=0.25,
**kwargs)
def faster_rcnn_resnet101_v1d_coco(pretrained=False,
pretrained_base=True,
**kwargs):
r"""Faster RCNN model from the paper
"Ren, S., He, K., Girshick, R., & Sun, J. (2015). Faster r-cnn: Towards
real-time object detection with region proposal networks"
Parameters
----------
pretrained : bool, optional, default is False
Load pretrained weights.
pretrained_base : bool or str, optional, default is True
Load pretrained base network, the extra layers are randomized. Note that
if pretrained is `True`, this has no effect.
ctx : Context, default CPU
The context in which to load the pretrained weights.
root : str, default '~/.mxnet/models'
Location for keeping the model parameters.
Examples
--------
>>> model = get_faster_rcnn_resnet101_v1d_coco(pretrained=True)
>>> print(model)
"""
from data.mscoco.detection_cv import COCODetection
classes = COCODetection.CLASSES
pretrained_base = False if pretrained else pretrained_base
outputs = [[12, 21], [13, 2]]
features, top_features = _parse_network('resnet101_v1d', outputs,
pretrained_base)
train_patterns = '|'.join(
['.*dense', '.*rpn', '.*down(2|3|4)_conv', '.*layers(2|3|4)_conv'])
return get_faster_rcnn(name='resnet101_v1d',
dataset='coco',
pretrained=pretrained,
features=features,
top_features=top_features,
classes=classes,
short=800,
max_size=1333,
train_patterns=train_patterns,
nms_thresh=0.5,
nms_topk=-1,
post_nms=-1,
roi_mode='align',
roi_size=(14, 14),
strides=16,
clip=4.42,
rpn_in_channel=1024,
rpn_channel=1024,
base_size=16,
scales=(2, 4, 8, 16, 32),
ratios=(0.5, 1, 2),
alloc_size=(128, 128),
rpn_nms_thresh=0.7,
rpn_train_pre_nms=12000,
rpn_train_post_nms=2000,
rpn_test_pre_nms=6000,
rpn_test_post_nms=1000,
rpn_min_size=0,
num_sample=128,
pos_iou_thresh=0.5,
pos_ratio=0.25,
max_num_gt=100,
**kwargs)