Spaces:
Running
on
A10G
Running
on
A10G
"""MobileNet and MobileNetV2.""" | |
''' | |
Code adopted from https://github.com/LikeLy-Journey/SegmenTron/blob/master/segmentron/models/backbones/mobilenet.py | |
''' | |
import torch | |
import torch.nn as nn | |
import torch.nn.functional as F | |
# ============ Basic Blocks ============ | |
class _ConvBNReLU(nn.Module): | |
def __init__(self, in_channels, out_channels, kernel_size, stride=1, padding=0, | |
dilation=1, groups=1, relu6=False, norm_layer=nn.BatchNorm2d): | |
super(_ConvBNReLU, self).__init__() | |
self.conv = nn.Conv2d(in_channels, out_channels, kernel_size, stride, padding, dilation, groups, bias=False) | |
self.bn = norm_layer(out_channels) | |
self.relu = nn.ReLU6(True) if relu6 else nn.ReLU(True) | |
def forward(self, x): | |
x = self.conv(x) | |
x = self.bn(x) | |
x = self.relu(x) | |
return x | |
class _DepthwiseConv(nn.Module): | |
"""conv_dw in MobileNet""" | |
def __init__(self, in_channels, out_channels, stride, norm_layer=nn.BatchNorm2d, **kwargs): | |
super(_DepthwiseConv, self).__init__() | |
self.conv = nn.Sequential( | |
_ConvBNReLU(in_channels, in_channels, 3, stride, 1, groups=in_channels, norm_layer=norm_layer), | |
_ConvBNReLU(in_channels, out_channels, 1, norm_layer=norm_layer)) | |
def forward(self, x): | |
return self.conv(x) | |
class InvertedResidual(nn.Module): | |
def __init__(self, in_channels, out_channels, stride, expand_ratio, dilation=1, norm_layer=nn.BatchNorm2d): | |
super(InvertedResidual, self).__init__() | |
assert stride in [1, 2] | |
self.use_res_connect = stride == 1 and in_channels == out_channels | |
layers = list() | |
inter_channels = int(round(in_channels * expand_ratio)) | |
if expand_ratio != 1: | |
# pw | |
layers.append(_ConvBNReLU(in_channels, inter_channels, 1, relu6=True, norm_layer=norm_layer)) | |
layers.extend([ | |
# dw | |
_ConvBNReLU(inter_channels, inter_channels, 3, stride, dilation, dilation, | |
groups=inter_channels, relu6=True, norm_layer=norm_layer), | |
# pw-linear | |
nn.Conv2d(inter_channels, out_channels, 1, bias=False), | |
norm_layer(out_channels)]) | |
self.conv = nn.Sequential(*layers) | |
def forward(self, x): | |
if self.use_res_connect: | |
return x + self.conv(x) | |
else: | |
return self.conv(x) | |
# ============ Backbone ============ | |
class MobileNetV2(nn.Module): | |
def __init__(self, num_classes=1000, norm_layer=nn.BatchNorm2d): | |
super(MobileNetV2, self).__init__() | |
output_stride = 8 | |
self.multiplier = 1 | |
if output_stride == 32: | |
dilations = [1, 1] | |
elif output_stride == 16: | |
dilations = [1, 2] | |
elif output_stride == 8: | |
dilations = [2, 4] | |
else: | |
raise NotImplementedError | |
inverted_residual_setting = [ | |
# t, c, n, s | |
[1, 16, 1, 1], | |
[6, 24, 2, 2], | |
[6, 32, 3, 2], | |
[6, 64, 4, 2], | |
[6, 96, 3, 1], | |
[6, 160, 3, 2], | |
[6, 320, 1, 1]] | |
# building first layer | |
input_channels = int(32 * self.multiplier) if self.multiplier > 1.0 else 32 | |
# last_channels = int(1280 * multiplier) if multiplier > 1.0 else 1280 | |
self.conv1 = _ConvBNReLU(3, input_channels, 3, 2, 1, relu6=True, norm_layer=norm_layer) | |
# building inverted residual blocks | |
self.planes = input_channels | |
self.block1 = self._make_layer(InvertedResidual, self.planes, inverted_residual_setting[0:1], | |
norm_layer=norm_layer) | |
self.block2 = self._make_layer(InvertedResidual, self.planes, inverted_residual_setting[1:2], | |
norm_layer=norm_layer) | |
self.block3 = self._make_layer(InvertedResidual, self.planes, inverted_residual_setting[2:3], | |
norm_layer=norm_layer) | |
self.block4 = self._make_layer(InvertedResidual, self.planes, inverted_residual_setting[3:5], | |
dilations[0], norm_layer=norm_layer) | |
self.block5 = self._make_layer(InvertedResidual, self.planes, inverted_residual_setting[5:], | |
dilations[1], norm_layer=norm_layer) | |
self.last_inp_channels = self.planes | |
# building last several layers | |
# features = list() | |
# features.append(_ConvBNReLU(input_channels, last_channels, 1, relu6=True, norm_layer=norm_layer)) | |
# features.append(nn.AdaptiveAvgPool2d(1)) | |
# self.features = nn.Sequential(*features) | |
# | |
# self.classifier = nn.Sequential( | |
# nn.Dropout2d(0.2), | |
# nn.Linear(last_channels, num_classes)) | |
# weight initialization | |
for m in self.modules(): | |
if isinstance(m, nn.Conv2d): | |
nn.init.kaiming_normal_(m.weight, mode='fan_out') | |
if m.bias is not None: | |
nn.init.zeros_(m.bias) | |
elif isinstance(m, nn.BatchNorm2d): | |
nn.init.ones_(m.weight) | |
nn.init.zeros_(m.bias) | |
elif isinstance(m, nn.Linear): | |
nn.init.normal_(m.weight, 0, 0.01) | |
if m.bias is not None: | |
nn.init.zeros_(m.bias) | |
def _make_layer(self, block, planes, inverted_residual_setting, dilation=1, norm_layer=nn.BatchNorm2d): | |
features = list() | |
for t, c, n, s in inverted_residual_setting: | |
out_channels = int(c * self.multiplier) | |
stride = s if dilation == 1 else 1 | |
features.append(block(planes, out_channels, stride, t, dilation, norm_layer)) | |
planes = out_channels | |
for i in range(n - 1): | |
features.append(block(planes, out_channels, 1, t, norm_layer=norm_layer)) | |
planes = out_channels | |
self.planes = planes | |
return nn.Sequential(*features) | |
def forward(self, x, side_feature): | |
x = self.conv1(x) | |
x = x + side_feature | |
x = self.block1(x) | |
c1 = self.block2(x) | |
c2 = self.block3(c1) | |
c3 = self.block4(c2) | |
c4 = self.block5(c3) | |
# x = self.features(x) | |
# x = self.classifier(x.view(x.size(0), x.size(1))) | |
return c1, c2, c3, c4 | |
def mobilenet_v2(norm_layer=nn.BatchNorm2d): | |
return MobileNetV2(norm_layer=norm_layer) | |
# ============ Segmentor ============ | |
class LRASPP(nn.Module): | |
"""Lite R-ASPP""" | |
def __init__(self, in_channels, out_channels, norm_layer=nn.BatchNorm2d, **kwargs): | |
super(LRASPP, self).__init__() | |
self.b0 = nn.Sequential( | |
nn.Conv2d(in_channels, out_channels, 1, bias=False), | |
norm_layer(out_channels), | |
nn.ReLU(True) | |
) | |
self.b1 = nn.Sequential( | |
nn.AdaptiveAvgPool2d((2,2)), | |
nn.Conv2d(in_channels, out_channels, 1, bias=False), | |
nn.Sigmoid(), | |
) | |
def forward(self, x): | |
size = x.size()[2:] | |
feat1 = self.b0(x) | |
feat2 = self.b1(x) | |
feat2 = F.interpolate(feat2, size, mode='bilinear', align_corners=True) | |
x = feat1 * feat2 | |
return x | |
class MobileSeg(nn.Module): | |
def __init__(self, nclass=1, **kwargs): | |
super(MobileSeg, self).__init__() | |
self.backbone = mobilenet_v2() | |
self.lraspp = LRASPP(320,128) | |
self.fusion_conv1 = nn.Conv2d(128,16,1,1,0) | |
self.fusion_conv2 = nn.Conv2d(24,16,1,1,0) | |
self.head = nn.Conv2d(16,nclass,1,1,0) | |
self.aux_head = nn.Conv2d(16,nclass,1,1,0) | |
def forward(self, x, side_feature): | |
x4, _, _, x8 = self.backbone(x, side_feature) | |
x8 = self.lraspp(x8) | |
x8 = F.interpolate(x8, x4.size()[2:], mode='bilinear', align_corners=True) | |
x8 = self.fusion_conv1(x8) | |
pred_aux = self.aux_head(x8) | |
x4 = self.fusion_conv2(x4) | |
x = x4 + x8 | |
pred = self.head(x) | |
return pred, pred_aux, x | |
def load_pretrained_weights(self, path_to_weights= ' '): | |
backbone_state_dict = self.backbone.state_dict() | |
pretrained_state_dict = torch.load(path_to_weights, map_location='cpu') | |
ckpt_keys = set(pretrained_state_dict.keys()) | |
own_keys = set(backbone_state_dict.keys()) | |
missing_keys = own_keys - ckpt_keys | |
unexpected_keys = ckpt_keys - own_keys | |
print('Loading Mobilnet V2') | |
print('Missing Keys: ', missing_keys) | |
print('Unexpected Keys: ', unexpected_keys) | |
backbone_state_dict.update(pretrained_state_dict) | |
self.backbone.load_state_dict(backbone_state_dict, strict= False) | |
class ScaleLayer(nn.Module): | |
def __init__(self, init_value=1.0, lr_mult=1): | |
super().__init__() | |
self.lr_mult = lr_mult | |
self.scale = nn.Parameter( | |
torch.full((1,), init_value / lr_mult, dtype=torch.float32) | |
) | |
def forward(self, x): | |
scale = torch.abs(self.scale * self.lr_mult) | |
return x * scale | |
# ============ Interactive Segmentor ============ | |
class BaselineModel(nn.Module): | |
def __init__(self, backbone_lr_mult=0.1, | |
norm_layer=nn.BatchNorm2d, **kwargs): | |
super().__init__() | |
self.feature_extractor = MobileSeg() | |
side_feature_ch = 32 | |
mt_layers = [ | |
nn.Conv2d(in_channels=3, out_channels=16, kernel_size=3, stride=2, padding=1), | |
nn.LeakyReLU(negative_slope=0.2), | |
nn.Conv2d(in_channels=16, out_channels=side_feature_ch, kernel_size=3, stride=1, padding=1), | |
ScaleLayer(init_value=0.05, lr_mult=1) | |
] | |
self.maps_transform = nn.Sequential(*mt_layers) | |
def backbone_forward(self, image, coord_features=None): | |
mask, mask_aux, feature = self.feature_extractor(image, coord_features) | |
return {'instances': mask, 'instances_aux':mask_aux, 'feature': feature} | |
def prepare_input(self, image): | |
prev_mask = torch.zeros_like(image)[:,:1,:,:] | |
return image, prev_mask | |
def forward(self, image, coarse_mask): | |
image, prev_mask = self.prepare_input(image) | |
coord_features = torch.cat((prev_mask, coarse_mask, coarse_mask * 0.0), dim=1) | |
click_map = coord_features[:,1:,:,:] | |
coord_features = self.maps_transform(coord_features) | |
outputs = self.backbone_forward(image, coord_features) | |
pred = nn.functional.interpolate( | |
outputs['instances'], | |
size=image.size()[2:], | |
mode='bilinear', align_corners=True | |
) | |
outputs['instances'] = torch.sigmoid(pred) | |
return outputs | |