#! /usr/bin/env python3
# -*- coding: utf-8 -*-
# File : vgg.py
# Author : Jiayuan Mao
# Email : maojiayuan@gmail.com
# Date : 03/31/2018
#
# This file is part of Jacinle.
# Distributed under terms of the MIT license.
"""Functions to build VGG models."""
import math
import functools
import torch.nn as nn
import torch.utils.model_zoo as model_zoo
from jactorch.io import load_state_dict
__all__ = [
'VGG', 'make_vgg',
'vgg11', 'vgg11_bn',
'vgg13', 'vgg13_bn',
'vgg16', 'vgg16_bn',
'vgg19', 'vgg19_bn',
'reset_vgg_parameters'
]
[docs]
class VGG(nn.Module):
[docs]
def __init__(self, cfg, batch_norm=False, incl_p5=True, incl_fcs=True, num_classes=1000):
super(VGG, self).__init__()
self.incl_p5 = incl_p5
self.incl_fcs = self.incl_p5 and incl_fcs
self.incl_cls = self.incl_fcs and num_classes is not None
if not self.incl_p5 and cfg[-1] == 'M':
cfg.pop()
self.features = self.make_layers(cfg, batch_norm=batch_norm)
if self.incl_fcs:
self.fc = nn.Sequential(
nn.Linear(512 * 7 * 7, 4096),
nn.ReLU(True),
nn.Dropout(),
nn.Linear(4096, 4096),
nn.ReLU(True),
nn.Dropout(),
)
if self.incl_cls:
self.classifier = nn.Linear(4096, num_classes)
self.reset_parameters()
[docs]
def reset_parameters(self):
reset_vgg_parameters(self)
[docs]
def forward(self, x):
x = self.features(x)
if self.incl_fcs:
x = x.view(x.size(0), -1)
x = self.fc(x)
if self.incl_cls:
x = self.classifier(x)
return x
[docs]
@staticmethod
def make_layers(cfg, batch_norm=False):
layers = []
in_channels = 3
for v in cfg:
if v == 'M':
layers += [nn.MaxPool2d(kernel_size=2, stride=2)]
else:
conv2d = nn.Conv2d(in_channels, v, kernel_size=3, padding=1)
if batch_norm:
layers += [conv2d, nn.BatchNorm2d(v), nn.ReLU(inplace=True)]
else:
layers += [conv2d, nn.ReLU(inplace=True)]
in_channels = v
return nn.Sequential(*layers)
cfgs = {
'A': [64, 'M', 128, 'M', 256, 256, 'M', 512, 512, 'M', 512, 512, 'M'],
'B': [64, 64, 'M', 128, 128, 'M', 256, 256, 'M', 512, 512, 'M', 512, 512, 'M'],
'D': [64, 64, 'M', 128, 128, 'M', 256, 256, 256, 'M', 512, 512, 512, 'M', 512, 512, 512, 'M'],
'E': [64, 64, 'M', 128, 128, 'M', 256, 256, 256, 256, 'M', 512, 512, 512, 512, 'M', 512, 512, 512, 512, 'M'],
}
model_urls = {
'vgg11': 'https://download.pytorch.org/models/vgg11-bbd30ac9.pth',
'vgg13': 'https://download.pytorch.org/models/vgg13-c768596a.pth',
'vgg16': 'https://download.pytorch.org/models/vgg16-397923af.pth',
'vgg19': 'https://download.pytorch.org/models/vgg19-dcbb9e9d.pth',
'vgg11_bn': 'https://download.pytorch.org/models/vgg11_bn-6002323d.pth',
'vgg13_bn': 'https://download.pytorch.org/models/vgg13_bn-abd245e5.pth',
'vgg16_bn': 'https://download.pytorch.org/models/vgg16_bn-6c64b313.pth',
'vgg19_bn': 'https://download.pytorch.org/models/vgg19_bn-c79401a0.pth',
}
fc_mapping = {
'fc.0.weight': 'classifier.0.weight',
'fc.0.bias': 'classifier.0.bias',
'fc.3.weight': 'classifier.3.weight',
'fc.3.bias': 'classifier.3.bias',
'classifier.weight': 'classifier.6.weight',
'classifier.bias': 'classifier.6.bias'
}
[docs]
def make_vgg(cfg_id, batch_norm, pretrained, url_id, incl_fcs=True, num_classes=1000):
model = VGG(cfgs[cfg_id], batch_norm=batch_norm, incl_fcs=incl_fcs, num_classes=num_classes)
if pretrained:
pretrained_model = model_zoo.load_url(model_urls[url_id])
for k, v in fc_mapping.items():
pretrained_model[k] = pretrained_model.pop(v)
if num_classes != 1000:
del pretrained_model['classifier.weight']
del pretrained_model['classifier.bias']
try:
load_state_dict(model, pretrained_model)
except KeyError:
pass # Intentionally ignore the key error.
return model
[docs]
def make_vgg_contructor(cfg_id, url_id, batch_norm=False):
func = functools.partial(make_vgg, cfg_id=cfg_id, url_id=url_id, batch_norm=batch_norm)
func.__name__ = url_id
func.__doc__ = url_id.replace('vgg', 'VGG ').replace('_bn', ' (with batch normalization)')
func.__doc__ += '(configuration: {})'.format(cfg_id)
return func
vgg11 = make_vgg_contructor('A', 'vgg11')
vgg13 = make_vgg_contructor('B', 'vgg13')
vgg16 = make_vgg_contructor('D', 'vgg16')
vgg19 = make_vgg_contructor('E', 'vgg19')
vgg11_bn = make_vgg_contructor('A', 'vgg11_bn', batch_norm=True)
vgg13_bn = make_vgg_contructor('B', 'vgg13_bn', batch_norm=True)
vgg16_bn = make_vgg_contructor('D', 'vgg16_bn', batch_norm=True)
vgg19_bn = make_vgg_contructor('E', 'vgg19_bn', batch_norm=True)
[docs]
def reset_vgg_parameters(m, fc_std=0.01, bfc_std=0.001):
if isinstance(m, nn.Conv2d):
n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
m.weight.data.normal_(0, math.sqrt(2. / n))
if m.bias is not None:
m.bias.data.zero_()
elif isinstance(m, nn.BatchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()
elif isinstance(m, nn.Linear):
m.weight.data.normal_(0, fc_std)
if m.bias is not None:
m.bias.data.zero_()
elif isinstance(m, nn.Bilinear):
m.weight.data.normal_(0, bfc_std)
if m.bias is not None:
m.bias.data.zero_()
else:
for sub in m.modules():
if m != sub:
reset_vgg_parameters(sub, fc_std=fc_std, bfc_std=bfc_std)