def init_modules(self):
self.feature_extractor = ResNetFeatureExtractor(
self.feature_extractor_config)
self.modify_feature_extractor()
self.rpn_model = RPNModel(self.rpn_config)
if self.pooling_mode == 'align':
self.rcnn_pooling = RoIAlignAvg(self.pooling_size,
self.pooling_size, 1.0 / 16.0)
elif self.pooling_mode == 'ps':
self.rcnn_pooling = PSRoIPool(7, 7, 1.0 / 16, 7, self.n_classes)
elif self.pooling_mode == 'psalign':
raise NotImplementedError('have not implemented yet!')
elif self.pooling_mode == 'deformable_psalign':
raise NotImplementedError('have not implemented yet!')
self.mask_rcnn_pooling = RoIAlignAvg(14, 14, 1.0 / 16.0)
# self.rcnn_cls_pred = nn.Linear(2048, self.n_classes)
self.rcnn_cls_pred = nn.Conv2d(2048, self.n_classes, 3, 1, 1)
if self.reduce:
in_channels = 2048
else:
in_channels = 2048 * 4 * 4
if self.class_agnostic:
self.rcnn_bbox_pred = nn.Linear(in_channels, 4)
else:
self.rcnn_bbox_pred = nn.Linear(in_channels, 4 * self.n_classes)
# loss module
if self.use_focal_loss:
self.rcnn_cls_loss = FocalLoss(2)
else:
self.rcnn_cls_loss = functools.partial(F.cross_entropy,
reduce=False)
self.rcnn_kp_loss = functools.partial(F.cross_entropy,
reduce=False,
ignore_index=-1)
self.rcnn_bbox_loss = nn.modules.SmoothL1Loss(reduce=False)
# some 3d statistic
# some 2d points projected from 3d
self.rcnn_3d_pred = nn.Linear(in_channels, 3)
# self.rcnn_3d_loss = MultiBinLoss(num_bins=self.num_bins)
# self.rcnn_3d_loss = MultiBinRegLoss(num_bins=self.num_bins)
self.rcnn_3d_loss = OrientationLoss(split_loss=True)
self.keypoint_predictor = KeyPointPredictor2(1024)
def __init__(self, classes, class_agnostic, alpha_con=None):
super(_fasterRCNN, self).__init__()
self.classes = classes
self.n_classes = len(classes)
self.class_agnostic = class_agnostic
# crowds alpha_con
self.alpha_con = alpha_con
# label source
self.label_source = cfg.LABEL_SOURCE
# loss
self.RCNN_loss_cls = 0
self.RCNN_loss_bbox = 0
# define rpn
self.RCNN_rpn = _RPN(self.dout_base_model, self.classes,
self.n_classes)
self.RCNN_proposal_target = _ProposalTargetLayer(self.n_classes)
self.RCNN_roi_pool = _RoIPooling(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.RCNN_roi_align = RoIAlignAvg(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.RCNN_aggregation_layer = _RCNNAggregationLayer()
self.grid_size = cfg.POOLING_SIZE * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE
self.RCNN_roi_crop = _RoICrop()
def init_modules(self):
self.feature_extractor = ResNetFeatureExtractor(
self.feature_extractor_config)
self.rpn_model = IoURPNModel(self.rpn_config)
self.rcnn_pooling = RoIAlignAvg(self.pooling_size, self.pooling_size,
1.0 / 16.0)
self.rcnn_cls_pred = nn.Conv2d(2048, self.n_classes, 3, 1, 1)
in_channels = 2048
self.rcnn_iou = nn.Linear(in_channels, self.n_classes)
self.rcnn_iog = nn.Linear(in_channels, self.n_classes)
self.rcnn_iod = nn.Linear(in_channels, self.n_classes)
if self.class_agnostic:
self.rcnn_bbox_pred = nn.Linear(in_channels, 4)
else:
self.rcnn_bbox_pred = nn.Linear(in_channels, 4 * self.n_classes)
# loss module
if self.use_focal_loss:
self.rcnn_cls_loss = FocalLoss(2)
else:
self.rcnn_cls_loss = functools.partial(F.cross_entropy,
reduce=False)
self.rcnn_iou_loss = nn.MSELoss(reduce=False)
self.rcnn_bbox_loss = nn.modules.SmoothL1Loss(reduce=False)
def __init__(self,
classes,
class_agnostic,
meta_train,
meta_test=None,
meta_loss=None):
super(_fasterRCNN, self).__init__()
self.classes = classes
self.n_classes = len(classes)
self.class_agnostic = class_agnostic
self.meta_train = meta_train
self.meta_test = meta_test
self.meta_loss = meta_loss
# loss
self.RCNN_loss_cls = 0
self.RCNN_loss_bbox = 0
# define rpn
self.RCNN_rpn = _RPN(self.dout_base_model)
self.RCNN_proposal_target = _ProposalTargetLayer(self.n_classes)
self.RCNN_roi_pool = _RoIPooling(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.RCNN_roi_align = RoIAlignAvg(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.grid_size = cfg.POOLING_SIZE * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE
self.RCNN_roi_crop = _RoICrop()
def __init__(self, classes, class_agnostic):
super(_fasterRCNN2Fc, self).__init__()
self.classes = classes
self.n_classes = len(classes)
self.class_agnostic = class_agnostic
# loss
self.RCNN_loss_cls = 0
self.RCNN_loss_bbox = 0
# define rpn
self.RCNN_rpn = _RPN(self.dout_base_model)
self.RCNN_proposal_target = _ProposalTargetLayer(self.n_classes)
self.RCNN_roi_pool = _RoIPooling(
cfg.POOLING_SIZE, cfg.POOLING_SIZE, 1.0/16.0)
self.RCNN_roi_align = RoIAlignAvg(
cfg.POOLING_SIZE, cfg.POOLING_SIZE, 1.0/16.0)
self.RFCN_psroi_pool = None
self.grid_size = cfg.POOLING_SIZE * \
2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE
self.RCNN_roi_crop = _RoICrop()
# attention
# self.conv_new_1 = nn.Conv2d(2048, 256, 1)
# self.fc_new_1 = nn.dense(name='fc1_new_1', num_hidden=1024)
# self.fc_new_2 = nn.dense(name='fc2_new_2', num_hidden=1024)
self.fc1 = nn.Linear(2048, 1024)
self.fc2 = nn.Linear(1024, 1024)
self.nongt_dim = 256 if self.training else cfg.TEST.RPN_POST_NMS_TOP_N
def __init__(self, classes, class_agnostic, num_class=20):
super(_FPN, self).__init__()
self.num_classes = num_class
self.classes = classes
self.n_classes = len(classes)
self.class_agnostic = class_agnostic
# loss
self.RCNN_loss_cls = 0
self.RCNN_loss_bbox = 0
self.maxpool2d = nn.MaxPool2d(1, stride=2)
# define rpn
self.RCNN_rpn = _RPN_FPN(self.dout_base_model)
self.RCNN_proposal_target = _ProposalTargetLayer(self.n_classes)
# NOTE: the original paper used pool_size = 7 for cls branch, and 14 for mask branch, to save the
# computation time, we first use 14 as the pool_size, and then do stride=2 pooling for cls branch.
self.RCNN_roi_pool = _RoIPooling(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.RCNN_roi_align = RoIAlignAvg(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.grid_size = cfg.POOLING_SIZE * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE
self.RCNN_roi_crop = _RoICrop()
self.top = nn.Sequential(
nn.Linear(in_features=12544, out_features=4096, bias=True),
nn.ReLU(inplace=True), nn.Dropout(0.5),
nn.Linear(in_features=4096, out_features=4096, bias=True),
nn.ReLU(inplace=True), nn.Dropout(0.5))
self.fc8c = nn.Linear(4096, self.num_classes)
self.fc8d = nn.Linear(4096, self.num_classes)
def __init__(self,
classes,
class_agnostic,
shrink=1,
mimic=False,
rois=None):
super(_fasterRCNN, self).__init__()
self.shrink = shrink
self.student = True if shrink >= 2 else False
self.mimic = mimic
self.classes = classes
self.n_classes = len(classes)
self.class_agnostic = class_agnostic
# loss
self.RCNN_loss_cls = 0
self.RCNN_loss_bbox = 0
# define rpn
self.RCNN_rpn = _RPN(self.dout_base_model)
self.RCNN_proposal_target = _ProposalTargetLayer(self.n_classes)
self.RCNN_roi_pool = _RoIPooling(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.RCNN_roi_align = RoIAlignAvg(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.grid_size = cfg.POOLING_SIZE * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE
self.RCNN_roi_crop = _RoICrop()
def __init__(self, classes, class_agnostic):
super(_FPN, self).__init__()
self.classes = classes
self.n_classes = len(classes)
self.class_agnostic = class_agnostic
# loss
self.RCNN_loss_cls = 0
self.RCNN_loss_bbox = 0
self.maxpool2d = nn.MaxPool2d(1, stride=2)
# define rpn
self.RCNN_rpn = _RPN_FPN(self.dout_base_model)
self.RCNN_proposal_target = _ProposalTargetLayer(self.n_classes)
# NOTE: the original paper used pool_size = 7 for cls branch, and 14 for mask branch, to save the
# computation time, we first use 14 as the pool_size, and then do stride=2 pooling for cls branch.
self.RCNN_roi_pool = _RoIPooling(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.RCNN_roi_align = RoIAlignAvg(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.RCNN_roi_prroi = PrRoIPool2D(
cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0) ######## add precision roi pooling #####
self.grid_size = cfg.POOLING_SIZE * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE
self.RCNN_roi_crop = _RoICrop()
def __init__(self, classes, class_agnostic, lighthead=False, compact_mode=False):
super(_fasterRCNN, self).__init__()
self.classes = classes
self.n_classes = len(classes)
self.class_agnostic = class_agnostic
self.lighthead = lighthead
# loss
self.RCNN_loss_cls = 0
self.RCNN_loss_bbox = 0
# define Large Separable Convolution Layer
if self.lighthead:
self.lh_mode = 'S' if compact_mode else 'L'
self.lsconv = LargeSeparableConv2d(
self.dout_lh_base_model, bias=False, bn=False, setting=self.lh_mode)
self.lh_relu = nn.ReLU(inplace=True)
# define rpn
self.RCNN_rpn = _RPN(self.dout_base_model)
self.RCNN_proposal_target = _ProposalTargetLayer(self.n_classes)
self.RCNN_roi_pool = _RoIPooling(cfg.POOLING_SIZE, cfg.POOLING_SIZE, 1.0 / 16.0)
self.RCNN_roi_align = RoIAlignAvg(cfg.POOLING_SIZE, cfg.POOLING_SIZE, 1.0 / 16.0)
self.grid_size = cfg.POOLING_SIZE * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE
self.RCNN_roi_crop = _RoICrop()
self.rpn_time = None
self.pre_roi_time = None
self.roi_pooling_time = None
self.subnet_time = None
def __init__(self, classes, class_agnostic):
super(_fasterRCNN, self).__init__()
self.classes = classes
self.n_classes = len(classes)
self.class_agnostic = class_agnostic
# loss
self.RCNN_loss_cls = 0
self.RCNN_loss_bbox = 0
# define rpn
self.RCNN_rpn = _RPN(self.dout_base_model)
self.RCNN_proposal_target = _ProposalTargetLayer(self.n_classes)
self.RCNN_roi_pool = _RoIPooling(cfg.POOLING_SIZE, cfg.POOLING_SIZE, 1.0 / 16.0)
self.RCNN_roi_align = RoIAlignAvg(cfg.POOLING_SIZE, cfg.POOLING_SIZE, 1.0 / 16.0)
self.RCNN_deform_roi_pool_1 = DeformRoIFunction(pool_height=7, pool_width=7, spatial_scale=1.0 / 16.0,
no_trans=True, trans_std=0.1,
sample_per_part=4,
output_dim=256,
group_size=1, part_size=7)
self.RCNN_deform_roi_pool_2 = DeformRoIFunction(pool_height=7, pool_width=7, spatial_scale=1.0 / 16.0,
no_trans=False, trans_std=0.1,
sample_per_part=4,
output_dim=256,
group_size=1, part_size=7)
self.grid_size = cfg.POOLING_SIZE * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE
self.RCNN_roi_crop = _RoICrop()
def __init__(self, classes, class_agnostic):
super(_fasterRCNN, self).__init__()
self.classes = classes
self.n_classes = len(classes)
self.class_agnostic = class_agnostic
# loss
self.RCNN_loss_cls = 0
self.RCNN_loss_bbox = 0
# define rpn
self.RCNN_rpn = _RPN(self.dout_base_model)
self.RCNN_rpn_t = _RPN(self.dout_base_model)
self.RCNN_proposal_target = _ProposalTargetLayer(self.n_classes)
self.RCNN_roi_pool = _RoIPooling(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.RCNN_roi_align = RoIAlignAvg(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.grid_size = cfg.POOLING_SIZE * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE
self.RCNN_roi_crop = _RoICrop()
self.RCNN_imageDA_3 = _ImageDA(256)
self.RCNN_imageDA_4 = _ImageDA(512)
self.RCNN_imageDA = _ImageDA(self.dout_base_model)
self.RCNN_instanceDA = _InstanceDA()
def __init__(self, classes, class_agnostic):
super(CoupleNet, self).__init__()
self.classes = classes
self.n_classes = len(classes)
self.class_agnostic = class_agnostic
# loss
self.RCNN_loss_cls = 0
self.RCNN_loss_bbox = 0
self.box_num_classes = 1 if class_agnostic else self.n_classes
# define rpn
self.RCNN_rpn = _RPN(self.dout_base_model)
self.RCNN_proposal_target = _ProposalTargetLayer(self.n_classes)
self.RCNN_roi_pool = _RoIPooling(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.RCNN_roi_align = RoIAlignAvg(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.RCNN_roi_crop = _RoICrop()
self.RCNN_psroi_pool_cls = PSRoIPool(cfg.POOLING_SIZE,
cfg.POOLING_SIZE,
spatial_scale=1 / 16.0,
group_size=cfg.POOLING_SIZE,
output_dim=self.n_classes)
self.RCNN_psroi_pool_loc = PSRoIPool(cfg.POOLING_SIZE,
cfg.POOLING_SIZE,
spatial_scale=1 / 16.0,
group_size=cfg.POOLING_SIZE,
output_dim=self.box_num_classes *
4)
self.avg_pooling = nn.AvgPool2d(kernel_size=cfg.POOLING_SIZE,
stride=cfg.POOLING_SIZE)
self.grid_size = cfg.POOLING_SIZE * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE
def __init__(self, main_classes, sub_classes, class_agnostic):
super(_hierarchyFasterRCNN, self).__init__()
#self.classes = classes
self.main_classes = main_classes
self.sub_classes = sub_classes
self.n_sub_classes = len(sub_classes)
self.n_main_classes = len(main_classes)
self.class_agnostic = class_agnostic
# loss
self.RCNN_loss_cls = 0
self.RCNN_loss_bbox = 0
# define rpn
self.RCNN_rpn = _RPN(self.dout_base_model)
self.RCNN_proposal_target = _ProposalTargetLayer(self.n_sub_classes)
self.RCNN_roi_pool = _RoIPooling(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.RCNN_roi_align = RoIAlignAvg(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.RFCN_psroi_pool = None
self.grid_size = cfg.POOLING_SIZE * \
2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE
self.RCNN_roi_crop = _RoICrop()
def init_modules(self):
self.feature_extractor = ResNetFeatureExtractor(
self.feature_extractor_config)
self.rpn_model = RPNModel(self.rpn_config)
self.rcnn_pooling = RoIAlignAvg(self.pooling_size, self.pooling_size,
1.0 / 16.0)
# self.rcnn_cls_pred = nn.Conv2d(2048, self.n_classes, 3, 1, 1)
self.rcnn_cls_pred = nn.Linear(2048, self.n_classes)
self.rcnn_bbox_pred = nn.Conv2d(2048, 4, 3, 1, 1)
# if self.class_agnostic:
# self.rcnn_bbox_pred = nn.Linear(2048, 4)
# else:
# self.rcnn_bbox_pred = nn.Linear(2048, 4 * self.n_classes)
# loss module
# if self.use_focal_loss:
# self.rcnn_cls_loss = FocalLoss(2)
# else:
# self.rcnn_cls_loss = functools.partial(
# F.cross_entropy, reduce=False)
self.rcnn_cls_loss = nn.MSELoss(reduce=False)
self.rcnn_bbox_loss = nn.modules.SmoothL1Loss(reduce=False)
# attention
self.spatial_attention = nn.Conv2d(2048, 1, 3, 1, 1)
def _init_modules(self, load_model=True):
resnet = resnet101()
self.RCNN_base = nn.Sequential(resnet.conv1, resnet.bn1, resnet.relu,
resnet.maxpool, resnet.layer1,
resnet.layer2, resnet.layer3)
self.RCNN_top = nn.Sequential(resnet.layer4)
self.Linear_top = nn.Linear(2048, self.embedding_dim)
if load_model:
state_dict = torch.load(self.oneshot_model_path)['model']
self.load_state_dict({
k: v
for k, v in state_dict.items() if k in self.state_dict()
})
self.det_module = pseudo_siamese_det(self.det_model_path)
self.det_module.create_architecture(load_det_model=load_model)
self.det_module.training = False
self.RCNN_roi_crop = _RoICrop()
self.RCNN_roi_align = RoIAlignAvg(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.RCNN_roi_pool = _RoIPooling(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
def __init__(self, classes, num_ways, class_agnostic, meta_train, meta_test=None, meta_loss=None, transductive=None, visualization=None):
super(_fasterRCNN, self).__init__()
self.classes = classes
self.n_classes = len(classes)
self.class_agnostic = class_agnostic
self.meta_train = meta_train
self.meta_test = meta_test
self.meta_loss = meta_loss
self.simloss = True
self.dis_simloss = True
self.transductive = transductive
self.visualization = visualization
# loss
self.RCNN_loss_cls = 0
self.RCNN_loss_bbox = 0
# define rpn
self.RCNN_rpn = _RPN(self.dout_base_model)
self.RCNN_proposal_target = _ProposalTargetLayer(self.n_classes)
self.RCNN_roi_pool = _RoIPooling(cfg.POOLING_SIZE, cfg.POOLING_SIZE, 1.0 / 16.0)
self.RCNN_roi_align = RoIAlignAvg(cfg.POOLING_SIZE, cfg.POOLING_SIZE, 1.0 / 16.0)
self.grid_size = cfg.POOLING_SIZE * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE
self.RCNN_roi_crop = _RoICrop()
self.num_layers_g = 3
self.num_ways = num_ways
self.alpha = 0.5
def __init__(self, classes, class_agnostic):
super(_fasterRCNN_OCR, self).__init__()
self.classes = classes
self.n_classes = len(classes)
self.class_agnostic = class_agnostic
# loss
self.RCNN_loss_cls = 0
self.RCNN_loss_bbox = 0
# define rpn
self.RCNN_rpn = _RPN(self.dout_base_model)
self.RCNN_proposal_target = _ProposalTargetLayer(self.n_classes)
# self.RCNN_roi_pool = _RoIPooling(cfg.POOLING_SIZE, cfg.POOLING_SIZE, 1.0 / 16.0)
self.RCNN_roi_align = RoIAlignAvg(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.RCNN_ocr_roi_pooling = roi_pooling(2) # ocr roi_pooling
self.grid_size = cfg.POOLING_SIZE * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE
self.RCNN_roi_crop = _RoICrop()
# rnn初始化，隐藏节点256
nh = 256
nclass = len('0123456789.') + 1
self.rnn = nn.Sequential(BidirectionalLSTM(1024, nh, nh),
BidirectionalLSTM(nh, nh, nclass))
self.ctc_critition = CTCLoss().cuda()
def init_modules(self):
self.feature_extractor = ResNetFeatureExtractor(
self.feature_extractor_config)
self.rpn_model = RPNModel(self.rpn_config)
if self.pooling_mode == 'align':
self.rcnn_pooling = RoIAlignAvg(self.pooling_size,
self.pooling_size, 1.0 / 16.0)
elif self.pooling_mode == 'ps':
self.rcnn_pooling = PSRoIPool(7, 7, 1.0 / 16, 7, self.n_classes)
elif self.pooling_mode == 'psalign':
raise NotImplementedError('have not implemented yet!')
elif self.pooling_mode == 'deformable_psalign':
raise NotImplementedError('have not implemented yet!')
self.rcnn_cls_pred = nn.Linear(2048, self.n_classes * 2048)
# self.rcnn_cls_pred = nn.Conv2d(2048, self.n_classes, 3, 1, 1)
if self.class_agnostic:
# self.bottle_neck = nn.Sequential(
# nn.Linear(2048, 512),
# nn.BatchNorm2d(512),
# nn.ReLU(inplace=True),
# nn.Linear(512, 2048))
# self.rcnn_bbox_pred_top = nn.Linear(2048, 4)
# self.relu_top = nn.ReLU(inplace=True)
self.rcnn_bbox_pred = nn.Conv2d(2048, 4, 3, 1, 1)
else:
self.rcnn_bbox_pred = nn.Linear(2048, 4 * self.n_classes)
# loss module
if self.use_focal_loss:
self.rcnn_cls_loss = FocalLoss(2)
else:
self.rcnn_cls_loss = functools.partial(
F.cross_entropy, reduce=False)
self.rcnn_bbox_loss = nn.modules.SmoothL1Loss(reduce=False)
def __init__(self, args, num_classes=20):
super(WSDDN_VGG16, self).__init__()
self.args = args
self.num_classes = num_classes
self.pretrained_dir = os.path.join(args.dataroot, args.pretrained_path, args.pretrained_model)
# VGG16, pth(weakly게 아닐 수도 있음)
vgg_model = torchvision.models.vgg16()
# 찍어보자
if self.pretrained_dir is None:
logger.debug('There is no VGG16 pretrained model')
else:
logger.info('Loading pretrained VGG16')
state_dict = torch.load(self.pretrained_dir)
vgg_model.load_state_dict({k: v for k, v in state_dict.items() if k in vgg_model.state_dict()})
"""
Network debug
"""
for k in vgg_model.state_dict():
print('k, v', k)
#fc6 어디감?
self.base_network = nn.Sequential(*list(vgg_model.features._modules.values())[:-1])
self.top_network = nn.Sequential(*list(vgg_model.classifier._modules.values())[:-1])
self.fc8c = nn.Linear(4096, self.num_classes)
self.fc8d = nn.Linear(4096, self.num_classes)
# OICR 참고해보자
self.roi_pooling = _RoIPooling(7, 7, 1.0 / 16.0)
self.roi_align = RoIAlignAvg(7, 7, 1.0 / 16.0)
self._init_weights()
def __init__(self, classes, class_agnostic):
super(_fasterRCNN, self).__init__() # 继承父类的__init__()方法
self.classes = classes
self.n_classes = len(classes)
self.class_agnostic = class_agnostic
# loss
self.RCNN_loss_cls = 0
self.RCNN_loss_bbox = 0
# define rpn
# 实例化,RPN网络(self.dout_base_model)是512,vgg16子类中定义,输入rpn的维度
self.RCNN_rpn = _RPN(self.dout_base_model)
self.RCNN_proposal_target = _ProposalTargetLayer(self.n_classes)
self.RCNN_roi_pool = _RoIPooling(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.RCNN_roi_align = RoIAlignAvg(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
# grid_size = 7 * 2 = 14
self.grid_size = cfg.POOLING_SIZE * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE
self.RCNN_roi_crop = _RoICrop()
# dout_base_model = 512
self.RCNN_imageDA = _ImageDA(self.dout_base_model)
self.RCNN_instanceDA = _InstanceDA()
self.consistency_loss = torch.nn.MSELoss(size_average=False)
def __init__(self, classes, class_agnostic, rpn_batchsize):
super(_fasterRCNN, self).__init__()
self.classes = classes
self.n_classes = len(classes)
self.class_agnostic = class_agnostic
self.rpn_batchsize = rpn_batchsize
# loss
self.RCNN_loss_cls = 0
self.RCNN_loss_bbox = 0
# define rpn
self.RCNN_rpn = _RPN(self.dout_base_model, self.rpn_batchsize)
self.RCNN_proposal_target = _ProposalTargetLayer(self.n_classes)
print 'INFO: pooling size is: ', cfg.POOLING_SIZE_H, cfg.POOLING_SIZE_W
self.RCNN_roi_pool = _RoIPooling(cfg.POOLING_SIZE_H,
cfg.POOLING_SIZE_W, 1.0 / 16.0)
self.RCNN_roi_align = RoIAlignAvg(cfg.POOLING_SIZE_H,
cfg.POOLING_SIZE_W, 1.0 / 16.0)
## wrote by Xudong Wang
self.grid_size_H = cfg.POOLING_SIZE_H * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE_H
self.grid_size_W = cfg.POOLING_SIZE_W * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE_W
## end
self.grid_size = cfg.POOLING_SIZE * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE
self.RCNN_roi_crop = _RoICrop()
def __init__(self, classes, class_agnostic):
super(_fasterRCNN, self).__init__()
self.classes = classes
self.n_classes = len(classes)
self.class_agnostic = class_agnostic
# define rpn
self.RCNN_roi_pool = _RoIPooling(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.RCNN_roi_align = RoIAlignAvg(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.grid_size = cfg.POOLING_SIZE * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE
self.RCNN_roi_crop = _RoICrop()
self.spaCNN = SpaConv()
self.spa_cls_score = nn.Sequential(nn.Linear(5408, 1024),
nn.LeakyReLU(), nn.Dropout(p=0.5),
nn.Linear(1024, self.n_classes))
self.obj_cls_score = nn.Sequential(nn.Linear(300, 512), nn.LeakyReLU(),
nn.Linear(512, self.n_classes))
self.obj_attention = nn.Sequential(nn.Linear(300, 512), nn.LeakyReLU(),
nn.Linear(512, 6))
def __init__(self, classes, class_agnostic, transfer):
super(_fasterRCNN, self).__init__()
self.classes = classes
self.n_classes = len(classes)
self.class_agnostic = class_agnostic
# loss
self.RCNN_loss_cls = 0
self.RCNN_loss_bbox = 0
# define rpn
self.RCNN_rpn = _RPN(self.dout_base_model)
self.RCNN_proposal_target = _ProposalTargetLayer(self.n_classes)
self.RCNN_roi_pool = _RoIPooling(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.RCNN_roi_align = RoIAlignAvg(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.grid_size = cfg.POOLING_SIZE * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE
self.RCNN_roi_crop = _RoICrop()
#transfer setting
self.transfer = transfer
if self.transfer:
self.transfer_weight = Variable(torch.Tensor([cfg.TRANSFER_WEIGHT
]).cuda(),
requires_grad=True)
self.grl = cfg.TRANSFER_GRL
self.weight = Variable(torch.zeros(0).cuda(), requires_grad=False)
self.transfer_select = cfg.TRANSFER_SELECT
self.transfer_gamma = cfg.TRANSFER_GAMMA
def __init__(self, classes, class_agnostic, context, S_agent, T_agent, ts,
tt, select_num, candidate_num):
super(_fasterRCNN, self).__init__()
self.classes = classes
self.n_classes = len(classes)
self.class_agnostic = class_agnostic
self.select_num = select_num
self.candidate_num = candidate_num
print("self.select_num: %d self.candidate_num: %d " %
(self.select_num, self.candidate_num))
# loss
self.RCNN_loss_cls = 0
self.RCNN_loss_bbox = 0
self.context = context
# define rpn
self.RCNN_rpn = _RPN(self.dout_base_model)
self.RCNN_proposal_target = _ProposalTargetLayer(self.n_classes)
self.RCNN_roi_pool = _RoIPooling(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.RCNN_roi_align = RoIAlignAvg(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.epsilon_by_epoch = lambda epoch_idx: cfg.epsilon_final + (cfg.epsilon_start - \
cfg.epsilon_final) * math.exp(-1. * epoch_idx / cfg.epsilon_decay)
self.iter_dqn = 0
self.epsilon_by_epoch_T = lambda epoch_idx: cfg.epsilon_final + (cfg.epsilon_start - \
cfg.epsilon_final) * math.exp(-1. * epoch_idx / cfg.epsilon_decay)
self.iter_dqn_T = 0
self.grid_size = cfg.POOLING_SIZE * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE
self.RCNN_roi_crop = _RoICrop()
def __init__(self, classes, class_agnostic):
super(_fasterRCNN, self).__init__()
self.classes = classes
self.n_classes = len(classes)
self.class_agnostic = class_agnostic
# loss
self.RCNN_loss_cls = 0
self.RCNN_loss_bbox = 0
# define rpn
self.RCNN_rpn = _RPN(self.dout_base_model)
#for p in self.RCNN_rpn.parameters(): p.requires_grad = False
self.RCNN_proposal_target = _ProposalTargetLayer(self.n_classes)
self.RCNN_roi_pool = _RoIPooling(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
# roi pooling in vgg16
# stride = 16
self.RCNN_roi_align = RoIAlignAvg(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
# roi pooling in Generator Network
# stride = 2
self.RCNN_roi_pool_conv1 = _RoIPooling(cfg.POOLING_SIZE,
cfg.POOLING_SIZE, 1.0 / 2.0)
self.grid_size = cfg.POOLING_SIZE * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE
self.RCNN_roi_crop = _RoICrop()
def init_modules(self):
self.feature_extractor = ResNetFeatureExtractor(
self.feature_extractor_config)
self.rpn_model = RPNModel(self.rpn_config)
if self.pooling_mode == 'align':
self.rcnn_pooling = RoIAlignAvg(self.pooling_size,
self.pooling_size, 1.0 / 16.0)
elif self.pooling_mode == 'ps':
self.rcnn_pooling = PSRoIPool(7, 7, 1.0 / 16, 7, self.n_classes)
elif self.pooling_mode == 'psalign':
raise NotImplementedError('have not implemented yet!')
elif self.pooling_mode == 'deformable_psalign':
raise NotImplementedError('have not implemented yet!')
self.rcnn_cls_pred = nn.Linear(2048, self.n_classes)
if self.reduce:
in_channels = 2048
else:
in_channels = 2048 * 4 * 4
if self.class_agnostic:
self.rcnn_bbox_pred = nn.Linear(in_channels, 4)
else:
self.rcnn_bbox_pred = nn.Linear(in_channels, 4 * self.n_classes)
# loss module
if self.use_focal_loss:
self.rcnn_cls_loss = FocalLoss(2)
else:
self.rcnn_cls_loss = functools.partial(F.cross_entropy,
reduce=False)
self.rcnn_bbox_loss = nn.modules.SmoothL1Loss(reduce=False)
def __init__(self, classes, class_agnostic):
super(_Deconv, self).__init__()
self.classes = classes
self.n_classes = len(classes)
self.class_agnostic = class_agnostic
#loss
self.RCNN_loss_cls = 0
self.RCNN_loss_bbox = 0
self.maxpool2d = nn.MaxPool2d(1, stride=2)
#define rpn
# if USE_ONE_FEATURE == 0:
self.RCNN_rpn = _RPN_Deconv(self.dout_base_model)
# else:
# self.RCNN_rpn = _RPN(self.dout_base_model)
self.RCNN_proposal_target = _ProposalTargetLayer(self.n_classes)
self.RCNN_roi_pool = _RoIPooling(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.RCNN_roi_align = RoIAlignAvg(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.grid_size = cfg.POOLING_SIZE * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE
self.RCNN_roi_crop = _RoICrop()
def init_modules(self):
self.feature_extractor = ResNetFeatureExtractor(
self.feature_extractor_config)
self.rpn_model = RPNModel(self.rpn_config)
if self.pooling_mode == 'align':
self.rcnn_pooling = RoIAlignAvg(self.pooling_size,
self.pooling_size, 1.0 / 16.0)
elif self.pooling_mode == 'ps':
self.rcnn_pooling = PSRoIPool(7, 7, 1.0 / 16, 7, self.n_classes)
elif self.pooling_mode == 'psalign':
raise NotImplementedError('have not implemented yet!')
elif self.pooling_mode == 'deformable_psalign':
raise NotImplementedError('have not implemented yet!')
# self.rcnn_cls_pred = nn.Linear(2048, self.n_classes)
self.rcnn_cls_pred = nn.Conv2d(2048, self.n_classes, 3, 1, 1)
if self.reduce:
in_channels = 2048
else:
in_channels = 2048 * 4 * 4
if self.class_agnostic:
self.rcnn_bbox_pred = nn.Linear(in_channels, 4)
else:
self.rcnn_bbox_pred = nn.Linear(in_channels, 4 * self.n_classes)
# loss module
if self.use_focal_loss:
self.rcnn_cls_loss = FocalLoss(2)
else:
self.rcnn_cls_loss = functools.partial(F.cross_entropy,
reduce=False)
self.rcnn_bbox_loss = nn.modules.SmoothL1Loss(reduce=False)
# some 3d statistic
# some 2d points projected from 3d
# self.rcnn_3d_preds_new = nn.Linear(in_channels, 3 + 4 * self.num_bins)
self.rcnn_3d_loss = MultiBinLoss(num_bins=self.num_bins)
# dims
self.rcnn_dims_pred = nn.Sequential(
*[nn.Linear(in_channels, 256),
nn.ReLU(),
nn.Linear(256, 3)])
# angle
self.rcnn_angle_pred = nn.Sequential(*[
nn.Linear(in_channels, 256),
nn.ReLU(),
nn.Linear(256, self.num_bins * 2)
])
# angle conf
self.rcnn_angle_conf_pred = nn.Sequential(*[
nn.Linear(in_channels, 256),
nn.ReLU(),
nn.Linear(256, self.num_bins * 2)
])
def init_modules(self):
"""
some modules
"""
self.feature_extractor = OFTNetFeatureExtractor(
self.feature_extractor_config)
feats_reduce_1 = nn.Conv2d(128, self.feat_size, 1, 1, 0)
feats_reduce_2 = nn.Conv2d(256, self.feat_size, 1, 1, 0)
feats_reduce_3 = nn.Conv2d(512, self.feat_size, 1, 1, 0)
self.feats_reduces = nn.ModuleList(
[feats_reduce_1, feats_reduce_2, feats_reduce_3])
self.feat_collapse = nn.Conv2d(8, 1, 1, 1, 0)
self.rcnn_output_head = nn.Conv2d(1152, self.rcnn_output_channels, 1,
1, 0)
self.rpn_output_head = nn.Conv2d(256 * 4, self.rpn_output_channels, 1,
1, 0)
# loss
self.reg_loss = nn.L1Loss(reduce=False)
# self.reg_loss = nn.SmoothL1Loss(reduce=False)
# if self.use_focal_loss:
# self.conf_loss = FocalLoss(
# self.n_classes, alpha=0.2, gamma=2, auto_alpha=False)
# else:
# self.conf_loss = nn.CrossEntropyLoss(reduce=False)
self.conf_loss = nn.L1Loss(reduce=False)
self.angle_loss = MultiBinLoss(num_bins=self.num_bins)
# fusion layer
# self.upconv1 = nn.ConvTranspose2d(self.feat_size, self.feat_size, 2, 2,
# 0)
self.fusion1 = nn.Conv2d(2 * self.feat_size, self.feat_size, 3, 1, 1)
self.fusion1_bn = nn.BatchNorm2d(self.feat_size)
self.relu1 = nn.ReLU()
self.upconv2 = nn.ConvTranspose2d(self.feat_size, self.feat_size, 2, 2,
0)
self.upconv2_bn = nn.BatchNorm2d(self.feat_size)
self.upconv2_relu = nn.ReLU()
self.relu2 = nn.ReLU()
self.fusion2 = nn.Conv2d(2 * self.feat_size, self.feat_size, 3, 1, 1)
self.fusion2_bn = nn.BatchNorm2d(self.feat_size)
self.upconv3 = nn.ConvTranspose2d(self.feat_size, self.feat_size, 2, 2,
0)
self.upconv3_bn = nn.BatchNorm2d(self.feat_size)
self.upconv3_relu = nn.ReLU()
# self.fusion3 = nn.Conv2d(self.feat_size, self.feat_size, 3, 1, 1)
# self.relu3 = nn.ReLU()
self.rcnn_pooling = RoIAlignAvg(self.pooling_size, self.pooling_size,
1.0 / 8.0)
def __init__(self, classes, class_agnostic):
super(_fasterRCNN, self).__init__()
self.classes = classes
self.n_classes = len(classes)
self.class_agnostic = class_agnostic
# loss
self.RCNN_loss_cls = 0
self.RCNN_loss_bbox = 0
# define rpn
self.RCNN_rpn = _RPN(self.dout_base_model)
self.RCNN_proposal_target = _ProposalTargetLayer(self.n_classes)
self.RCNN_roi_pool = _RoIPooling(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.RCNN_roi_align = RoIAlignAvg(cfg.POOLING_SIZE, cfg.POOLING_SIZE,
1.0 / 16.0)
self.RCNN_roi_align_32 = RoIAlignAvg(cfg.POOLING_SIZE,
cfg.POOLING_SIZE, 1.0 / 32.0)
self.RCNN_roi_align_16 = RoIAlignAvg(14, 14, 1.0 / 16.0)
self.RCNN_roi_align_8 = RoIAlignAvg(28, 28, 1.0 / 8.0)
self.RCNN_roi_align_4 = RoIAlignAvg(56, 56, 1.0 / 4.0)
self.RCNN_roi_align_2 = RoIAlignAvg(112, 112, 1.0 / 2.0)
self.RCNN_roi_align_1 = RoIAlignAvg(224, 224, 1.0 / 1.0)
self.grid_size = cfg.POOLING_SIZE * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE
self.RCNN_roi_crop = _RoICrop()
