def __init__(self,
cfg: Dict,
traj_dim=256,
cont_dim=256,
latent_dim=128,
mode_dim=3,
v_dim=4):
super(CVAE, self).__init__()
self.Traj_Encoder = basicModel.Trajectory_Encoder(h_dim=traj_dim,
v_dim=v_dim)
self.Cont_Encoder = basicModel.Context_Encoder(cfg, cont_dim=cont_dim)
# 回归均值
self.encoder_mean = nn.Sequential(
nn.Linear(in_features=traj_dim + cont_dim, out_features=256),
nn.Linear(in_features=256, out_features=latent_dim),
# nn.BatchNorm1d(latent_dim)
)
# 回归方差
self.encoder_var = nn.Sequential(
nn.Linear(in_features=traj_dim + cont_dim, out_features=256),
nn.Linear(in_features=256, out_features=latent_dim),
# nn.BatchNorm1d(latent_dim)
)
self.future_len = cfg["model_params"]["future_num_frames"]
num_targets = 2 * self.future_len
self.num_preds = num_targets * mode_dim
self.num_modes = mode_dim
self.decoder_net = nn.Sequential(
nn.BatchNorm1d(latent_dim + cont_dim),
nn.Linear(in_features=latent_dim + cont_dim, out_features=2048),
nn.Linear(in_features=2048,
out_features=self.num_preds + mode_dim),
)
def __init__(self, cdf: Dict, h_dim=64, cont_dim=256):
super(discriminator, self).__init__()
self.encoder = basicModel.Trajectory_Encoder(h_dim=h_dim, v_dim=2)
self.classifier = nn.Sequential(
nn.Linear(in_features=h_dim + cont_dim, out_features=256),
nn.Linear(in_features=256, out_features=1), nn.LeakyReLU())
def __init__(self,
cnn_model='resnet34',
channels=3,
traj_dim=256,
cont_dim=256,
latent_dim=128,
mode_dim=3,
v_dim=4):
super(CVAE_single, self).__init__()
self.Traj_Encoder = basicModel.Trajectory_Encoder(h_dim=traj_dim,
v_dim=v_dim)
self.Cont_Encoder = basicModel.Context_Encoder(cnn_model=cnn_model,
channels=channels,
cont_dim=cont_dim)
# 回归均值
self.encoder_mean = nn.Sequential(
nn.Linear(in_features=traj_dim + cont_dim, out_features=256),
nn.Linear(in_features=256, out_features=latent_dim),
# nn.BatchNorm1d(latent_dim)
)
# 回归方差
self.encoder_var = nn.Sequential(
nn.Linear(in_features=traj_dim + cont_dim, out_features=256),
nn.Linear(in_features=256, out_features=latent_dim),
# nn.BatchNorm1d(latent_dim)
)
self.future_len = 12
num_targets = 2 * self.future_len
self.num_preds = num_targets
self.decoder_net = nn.Sequential(nn.BatchNorm1d(latent_dim +
cont_dim), )
def __init__(self, cfg: Dict, traj_dim=256, cont_dim=256, mode=3, v_dim=4):
super(generator, self).__init__()
self.Traj_Encoder = basicModel.Trajectory_Encoder(h_dim=traj_dim,
v_dim=v_dim)
self.Cont_Encoder = basicModel.Context_Encoder(cfg, cont_dim=cont_dim)
self.future_len = cfg["model_params"]["future_num_frames"]
self.num_modes = mode
self.num_preds = 2 * self.future_len * self.num_modes
self.gen = nn.Sequential(
nn.BatchNorm1d(traj_dim + cont_dim),
nn.Linear(in_features=traj_dim + cont_dim,
out_features=traj_dim + cont_dim),
nn.Linear(in_features=traj_dim + cont_dim,
out_features=self.num_preds + self.num_modes))
def __init__(self,
cnn_model='resnet34',
channels=3,
traj_dim=256,
cont_dim=256,
mode=3,
v_dim=2):
super(generator, self).__init__()
self.Traj_Encoder = basicModel.Trajectory_Encoder(h_dim=traj_dim,
v_dim=v_dim)
self.Cont_Encoder = basicModel.Context_Encoder(cnn_model='resnet34',
channels=3,
cont_dim=cont_dim)
self.future_len = 12
self.num_modes = mode
self.num_preds = 2 * self.future_len * self.num_modes
self.gen = nn.Sequential(
nn.BatchNorm1d(traj_dim + cont_dim),
nn.Linear(in_features=traj_dim + cont_dim, out_features=512),
nn.Linear(in_features=512, out_features=256),
nn.Linear(in_features=256,
out_features=self.num_preds + self.num_modes),
nn.LeakyReLU())