some modification

README.md

@@ -13,7 +13,7 @@ This repository provides the official implementation of our paper, "PhysORD: A N
 ## Dataset
 - This project utilizes the [TartanDrive](https://github.com/castacks/tartan_drive) dataset. Follow the instructions in its [repository](https://github.com/castacks/tartan_drive?tab=readme-ov-file#create-traintest-split) to create the `train`, `test-easy` and `test-hard` sets.
 - `test-easy` is used for validation during training, and `test-hard` for model evaluation.
-- We also provide pre-processed data for quick reproduction. Download the [train_val_easy_507_step20.pt / ...step5.pt](https://drive.google.com/drive/folders/16PX9j6SUU8_LB0vq5wj31WWVUY49cR8l?usp=sharing) file.
+- We also provide pre-processed data for quick reproduction. Download the [train_val_easy_507_step20.pt / ...step5.pt](https://drive.google.com/drive/folders/16PX9j6SUU8_LB0vq5wj31WWVUY49cR8l?usp=sharing) file into the [data](data) folder.
 
 ## Reproduce Guide
 To reproduce our result in the paper, you can follow the the steps below.

eval.py

@@ -5,7 +5,7 @@
 
 if __name__ == '__main__':
     parser = argparse.ArgumentParser()
-    parser.add_argument('--exp_name', default='physord_s5', type=str, help='experiment name')
+    parser.add_argument('--exp_name', default='physord', type=str, help='experiment name')
     parser.add_argument('--eval_data_fp', type=str, required=False, default='/data/data0/datasets/tartandrive/data/test-hard/', help='Path to test data')
     parser.add_argument('--timesteps', type=int, required=False, default=20, help='Number of timesteps to predict')
     parser.add_argument('--test_sample_interval', type=int, default=1, help='test data sample interval')
@@ -16,8 +16,8 @@
     # Load the model
     print("Loading the model ...")
     model = PhysORD(device=device, use_dVNet = True, time_step = 0.1).to(device)
-    model_dir = f'./result2/{args.exp_name}'
-    model_fp = f'{model_dir}/best/best-data507-timestep5.tar'
+    model_dir = f'./pretrained/{args.exp_name}'
+    model_fp = f'{model_dir}/best/best-data507-timestep20.tar'
     model.load_state_dict(torch.load(model_fp, map_location=device))
 
     # Load the data

physord/model.py

@@ -132,66 +132,31 @@ def step_forward(self, x):
             vk_next = vk_next[:,:,0]
 
             return torch.cat((qk_next, vk_next, omegak_next, sk, rpmk, uk), dim=1)
-
-    def predict_traininga(self, step_num, x, action):
-        # initial_x, initial_y = x[:, 0].clone(), x[:, 1].clone()
-        # x[:, 0] = x[:, 0] - initial_x
-        # x[:, 1] = x[:, 1] - initial_y
-
-        x = torch.cat((x, action[0:1,:]), dim = 1)
-        # print("x shape: ", x.shape)
-        xseq = x[None,:,:]
-        curx = x
-        for i in range(step_num):
-            nextx = self.step_forward(curx)
-            # curx = nextx
-            curx = torch.cat((nextx[:, :26], action[i+1:i+2,:]), dim = 1)
-            # print("curx shape: ", curx.shape)
-            xseq = torch.cat((xseq, curx[None,:,:]), dim = 0)
-        # for i in range(step_num + 1):
-        #     xseq[i, :, 0] = xseq[i, :, 0] + initial_x
-        #     xseq[i, :, 1] = xseq[i, :, 1] + initial_y
-
-        return xseq
 
     def efficient_evaluation(self, step_num, x, action):
-        # initial_x, initial_y = x[:, 0].clone(), x[:, 1].clone()
-        # x[:, 0] = x[:, 0] - initial_x
-        # x[:, 1] = x[:, 1] - initial_y
-
-        # x = torch.cat((x, action[0:1,:]), dim = 1)
-        # print("x shape: ", x.shape)
+        initial_x, initial_y = x[:, 0].clone(), x[:, 1].clone()
+        x[:, 0] = x[:, 0] - initial_x
+        x[:, 1] = x[:, 1] - initial_y
         xseq = x[None,:,:]
         curx = x
         for i in range(step_num):
             nextx = self.step_forward(curx)
-            # curx = nextx
             curx = torch.cat((nextx[:, :26], action[i+1, :, :]), dim = 1)
-            # print("curx shape: ", curx.shape)
             xseq = torch.cat((xseq, curx[None,:,:]), dim = 0)
-        # for i in range(step_num + 1):
-        #     xseq[i, :, 0] = xseq[i, :, 0] + initial_x
-        #     xseq[i, :, 1] = xseq[i, :, 1] + initial_y
+        for i in range(step_num + 1):
+            xseq[i, :, 0] = xseq[i, :, 0] + initial_x
+            xseq[i, :, 1] = xseq[i, :, 1] + initial_y
 
         return xseq
 
     def evaluation(self, step_num, traj):
-        # print("x shape: ", x.shape)
-        # print("traj: ", traj[:, 10])
         xseq = traj[0,:,:]
         xseq = xseq[None,:,:]
         curx = traj[0,:,:]
-        # print("curx shape: ", curx.shape)
         for i in range(step_num):
-            # print("curx", curx[10])
             nextx = self.step_forward(curx)
-            # curx = nextx
             curx = torch.cat((nextx[:, :26], traj[i+1, :,26:29]), dim = 1)
-            # print("curx shape: ", curx.shape)
             xseq = torch.cat((xseq, curx[None,:,:]), dim = 0)
-        # for i in range(step_num + 1):
-        #     xseq[i, :, 0] = xseq[i, :, 0] + initial_x
-        #     xseq[i, :, 1] = xseq[i, :, 1] + initial_y
 
         return xseq
 

train.py

@@ -137,11 +137,11 @@ def train(args, train_data, val_data):
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser(description=None)
-    parser.add_argument('--exp_name', default='physord_s5', type=str, help='experiment name')
+    parser.add_argument('--exp_name', default='physord', type=str, help='experiment name')
     parser.add_argument('--train_data_size', type=int, default=507, help='number of training data: 100% = 507, 80% = 406, 50% = 254, 10% = 51, 1%=5')
-    parser.add_argument('--timesteps', type=int, default=5, help='number of prediction steps')
-    parser.add_argument('--preprocessed_data_dir', default=None, type=str, help='directory of the preprocessed data.')
-    parser.add_argument('--save_dir', default="./result2/", type=str, help='where to save the trained model')
+    parser.add_argument('--timesteps', type=int, default=20, help='number of prediction steps')
+    parser.add_argument('--preprocessed_data_dir', default='./data/', type=str, help='directory of the preprocessed data.')
+    parser.add_argument('--save_dir', default="./result/", type=str, help='where to save the trained model')
     parser.add_argument('--val_sample_interval', type=int, default=1, help='validation_data')
     parser.add_argument('--early_stop', dest='early_stopping', action='store_true', help='early stopping?')
     parser.add_argument('--pretrained', default=None, type=str, help='Path to the pretrained model. If not provided, no pretrained model will be loaded.')

util/data_process.py

@@ -185,13 +185,6 @@ def get_test_data(eval_data_fp, norm_params, T, sample_intervals):
         state = get_state_seq_from_traj(traj, min_val_st, max_val_st, min_val_brake, max_val_brake)
         state = arrange_data_sample(state, T + 1, sample_intervals)
         state = torch.from_numpy(state) 
-
-        initial_x = state[0, :, 0].unsqueeze(0)
-        initial_y = state[0, :, 1].unsqueeze(0)
-        initial_x = initial_x.expand(state.size(0), -1)
-        initial_y = initial_y.expand(state.size(0), -1)
-        state[:, :, 0] = state[:, :, 0] - initial_x
-        state[:, :, 1] = state[:, :, 1] - initial_y
         states_test.append(state)
     test_x_cat = torch.cat(states_test, dim=1)
     return test_x_cat