def __init__(self):
super(ModelBuilder, self).__init__()
# build backbone
self.backbone = get_backbone(cfg.BACKBONE.TYPE, **cfg.BACKBONE.KWARGS)
# build adjust layer
if cfg.ADJUST.ADJUST:
self.neck = get_neck(cfg.ADJUST.TYPE, **cfg.ADJUST.KWARGS)
# build rpn head
if cfg.TRANSFORMER.TRANSFORMER:
assert (cfg.TRANSFORMER.KWARGS.hidden_dims ==
cfg.ADJUST.KWARGS.out_channels,
"AdjustLayer out_channels = hidden_dims")
self.tr_head = get_tr_head(cfg.TRANSFORMER.TYPE,
**cfg.TRANSFORMER.KWARGS)
self.criterion = SiamTrCriterion(cfg.TRAIN.CLS_WEIGHT,
cfg.TRAIN.LOC_WEIGHT,
cfg.TRAIN.IOU_WEIGHT)
else:
self.rpn_head = get_rpn_head(cfg.RPN.TYPE, **cfg.RPN.KWARGS)
# build mask head
if cfg.MASK.MASK:
self.mask_head = get_mask_head(cfg.MASK.TYPE,
**cfg.MASK.KWARGS)
if cfg.REFINE.REFINE:
self.refine_head = get_refine_head(cfg.REFINE.TYPE)
def __init__(self, cfg):
super(ModelBuilder, self).__init__()
self.cfg = cfg
# build backbone
self.backbone = get_backbone(cfg.BACKBONE.TYPE,
**cfg.BACKBONE.KWARGS)
# build adjust layer
if cfg.ADJUST.ADJUST:
self.neck = get_neck(cfg.ADJUST.TYPE,
**cfg.ADJUST.KWARGS)
# build rpn head
self.rpn_head = get_rpn_head(cfg.RPN.TYPE,
**cfg.RPN.KWARGS)
# build mask head
if cfg.MASK.MASK:
self.mask_head = get_mask_head(cfg.MASK.TYPE,
**cfg.MASK.KWARGS)
if cfg.REFINE.REFINE:
self.refine_head = get_refine_head(cfg.REFINE.TYPE)
def __init__(self):
super(ModelBuilder, self).__init__()
# build backbone
self.backbone = get_backbone(cfg.BACKBONE.TYPE, **cfg.BACKBONE.KWARGS)
# build adjust layer
if cfg.ADJUST.ADJUST:
self.neck = get_neck(cfg.ADJUST.TYPE, **cfg.ADJUST.KWARGS)
# build attention map
#self.att_mods_head = att_mods_head(cfg.atts.TYPE,
# **cfg.atts.KWARGS)
# build corner head
self.corner_mods_head = corners_mods_head(cfg.corners.TYPE,
**cfg.corners.KWARGS)
# build offs map
#self.off_mods_head = off_mods_head(cfg.offs.TYPE,
# **cfg.offs.KWARGS)
## build rpn head
#self.rpn_head = get_rpn_head(cfg.RPN.TYPE,
# **cfg.RPN.KWARGS)
# build mask head
if cfg.MASK.MASK:
self.mask_head = get_mask_head(cfg.MASK.TYPE, **cfg.MASK.KWARGS)
if cfg.REFINE.REFINE:
self.refine_head = get_refine_head(cfg.REFINE.TYPE)
def __init__(self):
super(ModelBuilder, self).__init__()
# build backbone
self.backbone = get_backbone(cfg.BACKBONE.TYPE, **cfg.BACKBONE.KWARGS)
# build adjust layer
if cfg.ADJUST.ADJUST:
self.neck = get_neck(cfg.ADJUST.TYPE, **cfg.ADJUST.KWARGS)
if cfg.TRACK.TYPE == 'SiamCARTracker':
# build car head
self.car_head = CARHead(cfg, 256)
# build response map
self.xcorr_depthwise = xcorr_depthwise
# build loss
self.loss_evaluator = make_siamcar_loss_evaluator(cfg)
self.down = nn.ConvTranspose2d(256 * 3, 256, 1, 1)
else:
# build rpn head
self.rpn_head = get_rpn_head(cfg.RPN.TYPE, **cfg.RPN.KWARGS)
# build mask head
if cfg.MASK.MASK:
self.mask_head = get_mask_head(cfg.MASK.TYPE, **cfg.MASK.KWARGS)
if cfg.REFINE.REFINE:
self.refine_head = get_refine_head(cfg.REFINE.TYPE)
def __init__(self):
super(ModelBuilder, self).__init__()
# build backbone
self.backbone = get_backbone(cfg.BACKBONE.TYPE, **cfg.BACKBONE.KWARGS)
# build adjust layer
if cfg.ADJUST.ADJUST:
self.neck = get_neck(cfg.ADJUST.TYPE, **cfg.ADJUST.KWARGS)
def get_latent_rpn_head(name, **kwargs):
RPNS = {'MultiLatentRPN': MultiLatentRPN}
return RPNS[name](**kwargs)
# build rpn head
if cfg.LATENT:
self.rpn_head = get_latent_rpn_head(cfg.RPN.TYPE, **cfg.RPN.KWARGS)
else:
self.rpn_head = get_rpn_head(cfg.RPN.TYPE, **cfg.RPN.KWARGS)
# build mask head
if cfg.MASK.MASK:
self.mask_head = get_mask_head(cfg.MASK.TYPE, **cfg.MASK.KWARGS)
if cfg.REFINE.REFINE:
self.refine_head = get_refine_head(cfg.REFINE.TYPE)
self.auto_params, self.auto_modules = self.get_layers_in_autocoder_model(
)
self.auto_opt = self.build_opt_autocoder()
# for nms label
self.score_size = (cfg.TRACK.INSTANCE_SIZE - cfg.TRACK.EXEMPLAR_SIZE) // \
cfg.ANCHOR.STRIDE + 1 + cfg.TRACK.BASE_SIZE
self.anchors = self.generate_anchor(self.score_size)
def __init__(self):
super(FrostTemplate, self).__init__()
self.backbone = get_backbone(cfg.BACKBONE.TYPE,
**cfg.BACKBONE.KWARGS)
self.neck = get_neck(cfg.ADJUST.TYPE,
**cfg.ADJUST.KWARGS)
self.rpn_head = get_rpn_head(cfg.RPN.TYPE,
**cfg.RPN.KWARGS)
def __init__(self):
super(Model, self).__init__()
# build backbone
self.backbone = get_backbone(cfg.BACKBONE.TYPE, **cfg.BACKBONE.KWARGS)
# build adjust layer
if cfg.ADJUST.ADJUST:
self.neck = get_neck(cfg.ADJUST.TYPE, **cfg.ADJUST.KWARGS)
# build rpn head
self.rpn_head = get_rpn_head(cfg.RPN.TYPE, **cfg.RPN.KWARGS)
def __init__(self):
super(ModelBuilder, self).__init__()
# build backbone
self.backbone = get_backbone(cfg.BACKBONE.TYPE,
**cfg.BACKBONE.KWARGS)
#是否添加gru模块
if cfg.GRU.USE_GRU:
self.grus =GRU_Model(cfg.GRU.SEQ_IN,cfg.GRU.SEQ_OUT)
if self.grus.seq_out_len !=1:
raise ValueError("For tracking task GRU_Model.seq_out_len must be set as 1\n",
"please check the value of __C.GRU.SEQ_OUT in config.py file"
)
self.zfs = [None] * self.grus.seq_in_len # 多帧模板图z的特征f,测试前向的时候使用
# build adjust layer （siamese rpn++才有这个层）
if cfg.ADJUST.ADJUST:
self.neck = get_neck(cfg.ADJUST.TYPE,
**cfg.ADJUST.KWARGS)
# build rpn head
self.rpn_head = get_rpn_head(cfg.RPN.TYPE,
**cfg.RPN.KWARGS)
# build mask head(siamese mask里面才有这一层)
if cfg.MASK.MASK:
self.mask_head = get_mask_head(cfg.MASK.TYPE,
**cfg.MASK.KWARGS)
if cfg.REFINE.REFINE:
self.refine_head = get_refine_head(cfg.REFINE.TYPE)
#---------------------tenosrboard监视用-------------------
# hanning窗口,为tensorboard显示预测结果提前出事
hanning = np.hanning(cfg.TRAIN.OUTPUT_SIZE) # 生成和输出特征图大小相同的hanning窗
window = np.outer(hanning, hanning) # 一维度的hanning窗口通过外积得到二位hanning 窗口
self.anchor_num = len(cfg.ANCHOR.RATIOS)*len(cfg.ANCHOR.SCALES)
self.window = torch.from_numpy(np.tile(window.flatten(), self.anchor_num)).cuda().float() #size*size*anchor_num展平
#按照给定的比例因子生成一个位置的多种anchor,输出shape为【anchor_num,4】其中输出的anchor的值为[-w * 0.5, -h * 0.5, w * 0.5, h * 0.5]
self.anchors = Anchors(cfg.ANCHOR.STRIDE, #8
cfg.ANCHOR.RATIOS, # [0.33, 0.5, 1, 2, 3]
cfg.ANCHOR.SCALES)
# 生成两种类型的anchor,第一种是左上右下坐标类型的，第二种是中心点类型的，shape均为【4，anchor_num,size,size]
self.anchors.generate_all_anchors(im_c=cfg.TRAIN.SEARCH_SIZE//2,
size=cfg.TRAIN.OUTPUT_SIZE)
self.anchors_tensor =torch.from_numpy(self.anchors.all_anchors[0]).cuda()
def __init__(self):
super(Model, self).__init__()
# build backbone
self.backbone = get_backbone(cfg.BACKBONE.TYPE, **cfg.BACKBONE.KWARGS)
# build adjust layer
if cfg.ADJUST.ADJUST:
self.neck = get_neck(cfg.ADJUST.TYPE, **cfg.ADJUST.KWARGS)
# build feature fusion
# self.feature_fusion = get_feature_fusion(cfg.FUSION.TYPE,
# **cfg.FUSION.KWARGS)
self.feature_fusion = xcorr_depthwise
# PolarMask head
self.mask_head = get_mask_head(cfg.MASK.TYPE, **cfg.MASK.KWARGS)
def __init__(self):
super(ModelBuilder, self).__init__()
# build backbone
self.backbone = get_backbone(cfg.BACKBONE.TYPE,
**cfg.BACKBONE.KWARGS)
# build car head
self.car_head = CARHead(cfg, 256)
# build response map
self.attention = Graph_Attention_Union(256, 256)
# build loss
self.loss_evaluator = make_siamcar_loss_evaluator(cfg)
def __init__(self):
super(Model, self).__init__()
# build backbone
self.backbone = get_backbone(cfg.BACKBONE.TYPE, **cfg.BACKBONE.KWARGS)
# build adjust layer
if cfg.ADJUST.ADJUST:
self.neck = get_neck(cfg.ADJUST.TYPE, **cfg.ADJUST.KWARGS)
# build fcos head
self.fcos_head = get_fcos_head(cfg.FCOS.TYPE, **cfg.FCOS.KWARGS)
# build response map
self.xcorr_depthwise = xcorr_depthwise
# build loss
self.loss_evaluator = make_fcos_loss_evaluator(cfg)
def __init__(self):
super(ModelBuilder, self).__init__()
# build backbone
self.backbone = get_backbone(cfg.BACKBONE.TYPE, **cfg.BACKBONE.KWARGS)
# build adjust layer
self.neck = get_neck(cfg.ADJUST.TYPE, **cfg.ADJUST.KWARGS)
# build non-local layer
# self.non_local = get_nonlocal(cfg.NONLOCAL.TYPE,
# **cfg.NONLOCAL.KWARGS)
# roi align for cropping center
self.roi_align = RoIAlign((7, 7), 1.0 / cfg.ANCHOR.STRIDE, 1)
# build rpn head
self.rpn_head = get_rpn_head(cfg.RPN.TYPE, **cfg.RPN.KWARGS)
def __init__(self):
super(ModelBuilder, self).__init__()
# build backbone
self.backbone = get_backbone(cfg.BACKBONE.TYPE, **cfg.BACKBONE.KWARGS)
# build adjust layer
if cfg.ADJUST.ADJUST:
self.neck = get_neck(cfg.ADJUST.TYPE, **cfg.ADJUST.KWARGS)
# build car head
self.car_head = CARHead(cfg, 256)
# build response map
self.xcorr_depthwise = xcorr_depthwise
# build loss
self.loss_evaluator = make_siamcar_loss_evaluator(cfg)
self.down = nn.ConvTranspose2d(256 * 3, 256, 1, 1)
def __init__(self):
super(ModelBuilder, self).__init__()
# build ContextBlock
self.ContextBlock = ContextBlock(256)
# build backbone
self.backbone = get_backbone(cfg.BACKBONE.TYPE, **cfg.BACKBONE.KWARGS)
# build adjust layer
if cfg.ADJUST.ADJUST:
#"AdjustAllLayer"
self.neck = get_neck(cfg.ADJUST.TYPE, **cfg.ADJUST.KWARGS)
# build rpn head
self.rpn_head = get_rpn_head(cfg.RPN.TYPE, **cfg.RPN.KWARGS)
# build mask head
if cfg.MASK.MASK:
self.mask_head = get_mask_head(cfg.MASK.TYPE, **cfg.MASK.KWARGS)
if cfg.REFINE.REFINE:
self.refine_head = get_refine_head(cfg.REFINE.TYPE)
def __init__(self):
super(ModelBuilder, self).__init__()
# build backbone
self.backbone = get_backbone(cfg.BACKBONE.TYPE, **cfg.BACKBONE.KWARGS)
# build adjust layer
if cfg.ADJUST.ADJUST:
self.neck = get_neck(cfg.ADJUST.TYPE, **cfg.ADJUST.KWARGS)
# build rpn head
self.rpn_head = get_rpn_head(cfg.RPN.TYPE, **cfg.RPN.KWARGS)
self.CUDA = torch.cuda.is_available()
# build mask head
if cfg.MASK.MASK:
self.mask_head = get_mask_head(cfg.MASK.TYPE, **cfg.MASK.KWARGS)
if cfg.REFINE.REFINE:
self.refine_head = get_refine_head(cfg.REFINE.TYPE)
self.zf = []
def __init__(self):
super(ModelBuilder, self).__init__()
# build backbone
self.backbone = get_backbone(cfg.BACKBONE.TYPE,
**cfg.BACKBONE.KWARGS)
# build adjust layer
'''from the function "template" and "track", it can be seen that self.neck is shared by X and Z'''
if cfg.ADJUST.ADJUST:
self.neck = get_neck(cfg.ADJUST.TYPE,
**cfg.ADJUST.KWARGS)
# build rpn head
self.rpn_head = get_rpn_head(cfg.RPN.TYPE,
**cfg.RPN.KWARGS)
# build mask head
if cfg.MASK.MASK:
self.mask_head = get_mask_head(cfg.MASK.TYPE,
**cfg.MASK.KWARGS)
if cfg.REFINE.REFINE:
self.refine_head = get_refine_head(cfg.REFINE.TYPE)
# from tqdm import tqdm
# from tqdm import trange
# pbar =tqdm(total=10000000)
# pbar.set_description("processing ")
# for i in range(10000000):
# pbar.update(1)
# pbar.close()
#
from pysot.core.config import cfg
from pysot.models.backbone import get_backbone
import torch
cfg.merge_from_file(
"/home/tempuser1/pysot/experiments/siamrpn_r50_l234_dwxcorr_gpu_visdrone/config.yaml"
)
backbone = get_backbone(cfg.BACKBONE.TYPE, **cfg.BACKBONE.KWARGS)
data = torch.rand([1, 3, 255, 255])
output = backbone.forward(data)
for i in output:
print(i.size())
