diff --git a/plot_fig3.py b/plot_fig3.py
new file mode 100644
index 0000000..f5a87a0
--- /dev/null
+++ b/plot_fig3.py
@@ -0,0 +1,125 @@
+import sys
+import argparse
+import numpy as np
+import torch
+from EigenTrajectory import TrajNorm
+from CurveModel import *
+from utils import *
+import matplotlib.pyplot as plt
+import os
+
+# Reproducibility
+reproducibility_settings(seed=0)
+
+parser = argparse.ArgumentParser()
+parser.add_argument('--obs_len', type=int, default=8)
+parser.add_argument('--pred_len', type=int, default=12)
+
+
+def eval_method(args):
+    data_set = 'datasets/' + args.dataset + '/'
+    print('Scene: {}'.format(args.dataset))
+
+    args.batch_size = 1e8
+    loader_test = get_dataloader(data_set, 'train', args.obs_len, args.pred_len, args.batch_size)
+    # loader_test = get_dataloader(data_set, 'val', args.obs_len, args.pred_len, args.batch_size)
+    # loader_test = get_dataloader(data_set, 'test', args.obs_len, args.pred_len, args.batch_size)
+
+    # Preprocessing
+    obs_traj = loader_test.dataset.obs_traj
+    pred_traj = loader_test.dataset.pred_traj
+
+    n_ped, t_obs, dim = obs_traj.shape
+    n_ped, t_pred, dim = pred_traj.shape
+
+    # Normalization
+    traj_norm = TrajNorm(ori=True, rot=True, sca=False)
+    traj_norm.calculate_params(obs_traj)##计算obs_traj最后一帧的位置和最后两帧所形成的方向
+
+    obs_traj_norm = traj_norm.normalize(obs_traj)
+    pred_traj_norm = traj_norm.normalize(pred_traj)
+
+    # Singular Value Decomposition
+    print('===Singular Value Decomposition===')
+    A = obs_traj_norm.reshape(n_ped, t_obs * dim).T
+    B = pred_traj_norm.reshape(n_ped, t_pred * dim).T
+
+
+
+    U_obs, S_obs, Vt_obs = torch.linalg.svd(A, full_matrices=False)
+    U_pred, S_pred, Vt_pred = torch.linalg.svd(B, full_matrices=False)
+
+
+    for k in range(6, 7):
+        # Low-rank approximation
+        U_obs_trunc = U_obs[:, :k]
+        U_pred_trunc = U_pred[:, :k]
+
+        if os.path.exists('Eigen_fig3')==False:##创建文件夹
+            os.makedirs('Eigen_fig3')
+            
+        # Define a list of colors to cycle through
+        colors = ['blue', 'orange','green','red', 'purple', 'brown']
+            
+        for i in range(k):## code of Figure 3
+            xy_u_pred =U_pred_trunc[:, i].numpy().reshape(-1,2)##
+            
+            # Choose a color for this iteration (cycle through the colors list)
+            color = colors[i % len(colors)]
+            
+            # X-Y
+            plt.figure()
+            plt.plot(xy_u_pred[:,0], xy_u_pred[:,1], label=f'k={i}', color=color)
+            plt.xlim(-0.5, 0.5)
+            plt.ylim(-0.5, 0.5)
+            plt.legend()
+            plt.savefig(f'Eigen_fig3/xy_u_pred_{i}.png')
+            plt.close()
+            
+            # T-X
+            plt.figure()
+            plt.plot(xy_u_pred[:,0], label=f'k={i},x', color=color)
+            plt.xlim(0, 12)
+            plt.ylim(-0.5, 0.5)
+            plt.legend()
+            plt.savefig(f'Eigen_fig3/x_u_pred_{i}.png')
+            plt.close()
+            
+            # T-Y
+            plt.figure()
+            plt.plot(xy_u_pred[:,1], label=f'k={i},y', color=color)
+            plt.xlim(0, 12)
+            plt.ylim(-0.5, 0.5)
+            plt.legend()
+            plt.savefig(f'Eigen_fig3/y_u_pred_{i}.png')
+            plt.close()
+            
+
+            
+        C_obs = U_obs_trunc.T @ A
+        C_pred = U_pred_trunc.T @ B
+
+        A_recon = U_obs_trunc @ C_obs
+        B_recon = U_pred_trunc @ C_pred
+
+
+        # Denormalization
+        obs_traj_norm_recon = A_recon.T.reshape(n_ped, t_obs, dim)
+        pred_traj_norm_recon = B_recon.T.reshape(n_ped, t_pred, dim)
+        obs_traj_recon = traj_norm.denormalize(obs_traj_norm_recon)
+        pred_traj_recon = traj_norm.denormalize(pred_traj_norm_recon)
+
+        print('k: {}'.format(k), end='\t')
+        print('num params: {}'.format(k), end='\t')
+        print('obs error: {:.4f}'.format((obs_traj_recon - obs_traj).norm(p=2, dim=-1).mean().item()), end='\t')
+        print('pred error: {:.4f}'.format((pred_traj_recon - pred_traj).norm(p=2, dim=-1).mean().item()))
+
+
+if __name__ == '__main__':
+    args = parser.parse_args()
+    # dataset_all = ["eth", "hotel", "univ", "zara1", "zara2"]
+    dataset_all = ["eth"]
+
+    for scene in dataset_all:
+        args.dataset = scene
+        eval_method(args)