Trajectory Prediction

Module: Computer Vision | Difficulty: Advanced

Social LSTM

Social Pooling

Displacement Error

import torch
import torch.nn as nn

class SocialLSTM(nn.Module):
    def __init__(self, input_dim=2, hidden_dim=128, output_dim=2):
        super().__init__()
        self.lstm = nn.LSTM(input_dim, hidden_dim, batch_first=True)
        self.fc = nn.Linear(hidden_dim, output_dim)
    
    def forward(self, obs_seq):
        _, (h, c) = self.lstm(obs_seq)
        future = []
        h_t = h.squeeze(0)
        for _ in range(20):
            h_t = self.lstm(self.fc(h_t).unsqueeze(1), (h_t.unsqueeze(0), c))[1][0].squeeze(0)
            future.append(self.fc(h_t))
        return torch.stack(future, dim=1)

Key Takeaways

• Social interactions significantly affect pedestrian trajectories
• LSTM-based models capture temporal dynamics
• ADE and FDE are standard evaluation metrics