def __init__(self, config, device='cuda:0'): super().__init__() self.num_joints = config.model.backbone.num_joints self.volume_aggregation_method = config.model.volume_aggregation_method # volume self.volume_softmax = config.model.volume_softmax self.volume_multiplier = config.model.volume_multiplier self.volume_size = config.model.volume_size self.cuboid_side = config.model.cuboid_side self.kind = config.model.kind # heatmap self.heatmap_softmax = config.model.heatmap_softmax self.heatmap_multiplier = config.model.heatmap_multiplier # modules config.model.backbone.alg_confidences = False config.model.backbone.vol_confidences = False self.backbone = pose_resnet.get_pose_net(config.model.backbone, device=device) for p in self.backbone.final_layer.parameters(): p.requires_grad = False self.process_features = nn.Sequential(nn.Conv2d(256, 32, 1)) self.volume_net = V2VModel(32, self.num_joints) self.inputImageShape = [384, 384] self.outputHeatmapShape = [96, 96]
def __init__(self, config): super().__init__() config.model.backbone.alg_confidences = False config.model.backbone.vol_confidences = False self.backbone = pose_resnet.get_pose_net(config.model.backbone) # self.return_confidences = config.backbone.return_confidences self.direct_optimization = config.model.direct_optimization
def __init__(self, config, device='cuda:0'): super().__init__() config.model.backbone.alg_confidences = False config.model.backbone.vol_confidences = False self.backbone = pose_resnet.get_pose_net(config.model.backbone, device=device) self.direct_optimization = config.model.direct_optimization
def __init__(self, config, device='cuda:0'): super().__init__() self.use_confidences = config.model.use_confidences if config.model.use_default_backbone: self.backbone = pose_resnet.get_pose_net(config.model.backbone, config.opt.batch_size, device=device) else: self.backbone = pose_hrnet.get_pose_net(device=device) self.heatmap_softmax = config.model.heatmap_softmax self.heatmap_multiplier = config.model.heatmap_multiplier
def __init__(self, config): super().__init__() self.use_confidences = config.model.use_confidences config.model.backbone.alg_confidences = False config.model.backbone.vol_confidences = False if self.use_confidences: config.model.backbone.alg_confidences = True self.backbone = pose_resnet.get_pose_net(config.model.backbone) self.heatmap_softmax = config.model.heatmap_softmax # true self.heatmap_multiplier = config.model.heatmap_multiplier
def __init__(self, config, device='cuda:0'): super().__init__() self.num_joints = config.model.backbone.num_joints self.volume_aggregation_method = config.model.volume_net.volume_aggregation_method # volume self.volume_softmax = config.model.volume_net.volume_softmax self.volume_multiplier = config.model.volume_net.volume_multiplier self.volume_size = config.model.volume_net.volume_size self.cuboid_size = config.model.volume_net.cuboid_size self.kind = config.model.kind self.use_gt_pelvis = config.model.volume_net.use_gt_pelvis # heatmap self.heatmap_softmax = config.model.heatmap_softmax self.heatmap_multiplier = config.model.heatmap_multiplier # transfer self.transfer_cmu_to_human36m = config.dataset.transfer_cmu_to_human36m # modules config.model.backbone.alg_confidences = False config.model.backbone.vol_confidences = False if self.volume_aggregation_method.startswith('conf'): config.model.backbone.vol_confidences = True self.backbone = pose_resnet.get_pose_net(config.model.backbone) self.use_feature = config.model.volume_net.use_feature_v2v if config.model.backbone.fix_weights: for p in self.backbone.parameters(): p.requires_grad = False elif self.use_feature: for p in self.backbone.final_layer.parameters(): p.requires_grad = False v2v_in_channel = self.num_joints if self.use_feature: self.process_features = nn.Sequential( nn.Conv2d(256, 32, 1) ) v2v_in_channel = 32 self.volume_net = V2VNet(v2v_in_channel, self.num_joints, config)
def __init__(self, config, device='cuda:0'): super().__init__() self.num_joints = config.model.backbone.num_joints self.volume_aggregation_method = config.model.volume_aggregation_method # volume self.volume_softmax = config.model.volume_softmax self.volume_multiplier = config.model.volume_multiplier self.volume_size = config.model.volume_size self.cuboid_side = config.model.cuboid_side self.kind = config.model.kind self.use_gt_pelvis = config.model.use_gt_pelvis # heatmap self.heatmap_softmax = config.model.heatmap_softmax self.heatmap_multiplier = config.model.heatmap_multiplier # transfer self.transfer_cmu_to_human36m = config.model.transfer_cmu_to_human36m if hasattr( config.model, "transfer_cmu_to_human36m") else False # modules config.model.backbone.alg_confidences = False config.model.backbone.vol_confidences = False if self.volume_aggregation_method.startswith('conf'): config.model.backbone.vol_confidences = True self.backbone = pose_resnet.get_pose_net(config.model.backbone, config.opt.batch_size, device=device) for p in self.backbone.final_layer.parameters(): p.requires_grad = False self.process_features = nn.Sequential(nn.Conv2d(256, 32, 1)) self.volume_net = V2VModel(32, self.num_joints)
def __init__(self, config, device='cpu'): super().__init__() self.num_joints = 17 self.volume_aggregation_method = "softmax" # volume self.volume_softmax = True self.volume_multiplier = 1.0 self.volume_size = 64 self.cuboid_side = 2500.0 self.kind = "mpii" self.use_gt_pelvis = False # heatmap self.heatmap_softmax = True self.heatmap_multiplier = 1.0 # transfer self.transfer_cmu_to_human36m = False # modules config.model.backbone.alg_confidences = False config.model.backbone.vol_confidences = False if self.volume_aggregation_method.startswith('conf'): config.model.backbone.vol_confidences = True self.backbone = pose_resnet.get_pose_net(config.model.backbone, device=device) for p in self.backbone.final_layer.parameters(): p.requires_grad = False self.process_features = nn.Sequential(nn.Conv2d(256, 32, 1)) self.volume_net = V2VModel(32, self.num_joints)