ECEN689_Project/decoder.py at main · Zahid8/ECEN689_Project · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
import torch
import torch.nn as nn


class TransformerEncoder(nn.TransformerEncoder):
    def __init__(self, encoder_layer, num_layers, norm=None):
        super(TransformerEncoder, self).__init__(
            encoder_layer=encoder_layer, num_layers=num_layers, norm=norm
        )

    def forward(
        self,
        src,
        src_mask=None,
        src_key_padding_mask=None,
    ):
        output = src

        for mod in self.layers:
            output = mod(
                    output, src_mask=src_mask, src_key_padding_mask=src_key_padding_mask
                )

        if self.norm is not None:
            output = self.norm(output)

        return output


class Decoder(nn.Module):
    def __init__(
        self,
        hidden_dim,
        num_heads,
        dim_feedforward,
        num_layers,
        hist_len,
        fut_len,
        fc_in_traj,
        predict_head_traj,
        dropout=0.1,
        activation="relu",
        num_future=20,
    ):
        super().__init__()
        self.hidden_dim = hidden_dim
        self.hist_len = hist_len
        self.fut_len = fut_len
        self.seq_len = self.hist_len + self.fut_len
        self.num_future = num_future
        decoder_layer = nn.TransformerEncoderLayer(
                d_model=hidden_dim,
                nhead=num_heads,
                dim_feedforward=dim_feedforward,
                dropout=dropout,
                activation=activation,
            )

        self.decoder = TransformerEncoder(decoder_layer, num_layers=num_layers)

        self.fc_in_traj = fc_in_traj
        self.predict_head_traj = predict_head_traj

    def forward(self, seq_feat, training=True):
        B, C, seq_len, D = seq_feat.shape
        seq_feat = seq_feat.reshape(B, C * seq_len, D)  # [B, total_seq_len, hidden_dim]
        B, total_seq_len, D = seq_feat.shape
        seq_feat = seq_feat.permute(1, 0, 2)  # [total_seq_len, B, hidden_dim]
        outputs = self.decoder(
            src=seq_feat,
        )  # [total_seq_len, B, hidden_dim]

        outputs = outputs.permute(1, 0, 2)  # [B, total_seq_len, hidden_dim]

        outputs = outputs[:, -self.fut_len :]

        out_traj = []
        for k in range(self.num_future):
            temp_out = self.predict_head_traj[k](outputs)
            out_traj.append(temp_out)
        out_traj = torch.stack(out_traj)  # [K_modal, B, total_seq_len, hidden_dim]
        outputs = out_traj.permute(1, 0, 2, 3)  # [B, K_modal, seq_len, 2]
        outputs = outputs[:, :, -self.fut_len :]

        return outputs  # [B, K_modal, fut_len, 2] or [B, K_modal, total_seq_len, 2]


if __name__ == "__main__":
    B, hidden_dim = 4, 256
    C = 3
    dim_feedforward = 512
    num_heads = 8
    num_layers = 4
    hist_len = 9
    fut_len = 12
    seq_len = hist_len + fut_len
    num_future = 20

    seq_feat = torch.randn(B, C, seq_len, hidden_dim)

    predict_head_traj = []
    # Multiple prediction heads for trajectory
    for _ in range(num_future):
        predict_head_traj.append(nn.Linear(hidden_dim, 2, bias=False))
    predict_head_traj = nn.ModuleList(predict_head_traj)

    decoder = Decoder(
        hidden_dim,
        num_heads,
        dim_feedforward,
        num_layers,
        hist_len,
        fut_len,
        fc_in_traj=nn.Linear(2, hidden_dim),
        num_future=num_future,
        predict_head_traj=predict_head_traj
    )

    output, output_pi = decoder(seq_feat)

    print("Output shape:", output.shape)  # Expected: [B, num_future, fut_len, 2]

    if output_pi is not None:
        print("Output pi shape:", output_pi.shape)