def create_efficientnet(model_name,
pretrained=True,
num_classes=1000,
norm_layer=torch.nn.BatchNorm2d,
expand_ratio=6,
group_dim=1):
""" Creates EfficientNet instance with the predefined parameters.
Parameters:
model_name: Name of the model.
pretrained: if true the network is initialized with pretrained weights.
norm_layer: The used normalization layer in the network. eg. torch.nn.Identity means no initialization.
expand_ratio: The used expand ratio in the blocks. Official EfficientNet uses 6
group_dim: Dimensionality of the depthwise convolution. Official EfficientNet uses 1.
Returns:
The initialized EfficientNet model.
"""
EfficientNet._check_model_name_is_valid(model_name)
blocks_args, global_params = get_model_params(model_name,
{"num_classes": num_classes})
# Change expand ratio
for idx in range(1, len(blocks_args)):
blocks_args[idx] = blocks_args[idx]._replace(expand_ratio=expand_ratio)
model = EfficientNet(blocks_args, global_params)
model.set_swish(memory_efficient=False)
if group_dim > 1:
replace_en_depthwise_conv(model, group_dim)
if not isinstance(norm_layer, torch.nn.BatchNorm2d):
replace_bn(model, norm_layer)
init_efficientnet(model)
if pretrained:
pretrained_model = EfficientNet.from_name(model_name)
load_modified_model_from_state(model, pretrained_model.state_dict())
return model
def TwoWayFPNBackbone():
out_channels = 256
override_params = {'num_classes': 1000}
paras = get_model_params('efficientnet-b5', override_params)
model = EfficientNet(paras[0], paras[1])
arr = model.return_sub()
fpn = FPN(arr, paras[0], paras[1], out_channels=out_channels)
return fpn
def __init__(
self,
num_classes=40,
neck='bnneck',
neck_feat='after',
model_path='/home/zhoumi/.cache/torch/checkpoints/efficientnet-b4-6ed6700e.pth'
):
super(efficient_baseline, self).__init__()
#1.4, 1.8, 380, 0.4
blocks_args, global_params = efficientnet(width_coefficient=1.4,
depth_coefficient=1.8,
dropout_rate=0.4,
image_size=380)
self.base = EfficientNet(blocks_args=blocks_args,
global_params=global_params)
self.base.load_param(model_path)
print('Loading pretrained ImageNet model......')
# self.gap = nn.AdaptiveAvgPool2d(1)
# self.gap = nn.AdaptiveMaxPool2d(1)
self.num_classes = num_classes
self.neck = neck
self.neck_feat = neck_feat
if self.neck == 'no':
self.classifier = nn.Linear(self.in_planes, self.num_classes)
# self.classifier = nn.Linear(self.in_planes, self.num_classes, bias=False) # new add by luo
# self.classifier.apply(weights_init_classifier) # new add by luo
elif self.neck == 'bnneck':
self.bottleneck = nn.BatchNorm1d(self.in_planes)
self.bottleneck.bias.requires_grad_(False) # no shift
self.classifier = nn.Linear(self.in_planes,
self.num_classes,
bias=False)
self.bottleneck.apply(weights_init_kaiming)
self.classifier.apply(weights_init_classifier)
def create_backbone(self, compound_coefficient):
#todo : load diffirent EfficientNet based on compound_coefficient
return EfficientNet()
def from_name(cls, model_name, override_params=None):
cls._check_model_name_is_valid(model_name)
blocks_args, global_params = get_model_params(model_name,
override_params)
return EfficientNet(blocks_args, global_params)
