Fix a serialization issue on tfrecord, add some adaptability to plot script.

.gitignore

@@ -0,0 +1,4 @@
+./output
+./output/*
+./data/*
+./data

plot_deform.py

@@ -8,7 +8,7 @@
 from absl import app
 from absl import flags
 
-from matplotlib import animation
+from matplotlib import animation, cm
 import matplotlib.pyplot as plt
 
 import math
@@ -18,26 +18,29 @@
 
 import torch
 
-root_dir = pathlib.Path(__file__).parent.resolve()
-output_dir = os.path.join(root_dir, 'output', 'deforming_plate')
-all_subdirs = [os.path.join(output_dir, d) for d in os.listdir(output_dir) if
-               os.path.isdir(os.path.join(output_dir, d))]
-latest_subdir = max(all_subdirs, key=os.path.getmtime)
-rollout_path = os.path.join(latest_subdir, 'rollout', 'rollout.pkl')
+
+os.environ['KMP_DUPLICATE_LIB_OK']='TRUE'
+FLAGS = flags.FLAGS
+
+flags.DEFINE_string('path_prefix', 'output/deforming_plate', 'root dir to the output files relative to this script.')
+flags.DEFINE_string('rollout_path', None, 'specific rollout path to plot. Will plot all if set to None.')
 
 
 def main(unused_argv):
-    path_prefix = 'E:\\meshgraphnets\\output\\deforming_plate\\'
-    path_suffix = 'rollout\\rollout.pkl'
-    rollout_paths = ['Sat-Feb-19-15-44-13-2022']
-    # path_prefix = '/home/kit/anthropomatik/sn2444/meshgraphnets/output/deforming_plate/'
-    # rollout_paths = ['Mon-Jan-31-05-04-38-2022/2', 'Mon-Jan-31-05-10-30-2022/2', 'Mon-Jan-31-05-20-38-2022/2', 'Mon-Jan-31-05-35-42-2022/2', 'Mon-Jan-31-05-39-05-2022/2', 'Mon-Jan-31-08-28-21-2022/2']
+    path_prefix = FLAGS.path_prefix
+    path_suffix = 'rollout.pkl'
+
+    if not FLAGS.rollout_path:
+        rollout_paths = [d.name for d in os.scandir(path_prefix) if d.is_dir()]
+    else:
+        rollout_paths = [FLAGS.rollout_path]
+
     for rollout_path in rollout_paths:
         run_path = os.path.join(path_prefix, rollout_path)
         all_subdirs = [os.path.join(run_path, d) for d in os.listdir(run_path) if
                        os.path.isdir(os.path.join(run_path, d))]
         save_path = max(all_subdirs, key=os.path.getmtime)
-        data_path = os.path.join(path_prefix, save_path, path_suffix)
+        data_path = os.path.join(save_path, path_suffix)
         print("Ploting run", save_path)
         with open(data_path, 'rb') as fp:
             rollout_data = pickle.load(fp)
@@ -83,17 +86,17 @@ def animate(num):
             original_pos = torch.squeeze(rollout_data[traj]['gt_pos'], dim=0)[step].to('cpu')
 
             faces = torch.squeeze(rollout_data[traj]['faces'], dim=0)[step].to('cpu')
-
-            ax_origin.plot_trisurf(original_pos[:, 0], original_pos[:, 1], faces, original_pos[:, 2], shade=True, alpha=0.3)
-            ax_pred.plot_trisurf(pos[:, 0], pos[:, 1], faces, pos[:, 2], shade=True, alpha=0.3)
+            ax_origin.plot_trisurf(original_pos[:, 0], original_pos[:, 1], faces, original_pos[:, 2], shade=False, alpha=0.3)
+            ax_pred.plot_trisurf(pos[:, 0], pos[:, 1], faces, pos[:, 2], shade=False, alpha=0.3)
 
             ax_origin.set_title('ORIGIN Trajectory %d Step %d' % (traj, step))
             ax_pred.set_title('PRED Trajectory %d Step %d' % (traj, step))
             return fig,
 
         anima = animation.FuncAnimation(fig, animate, frames=math.floor(num_frames * 10), interval=100)
-        # writervideo = animation.FFMpegWriter(fps=30)
-        # anima.save(os.path.join(save_path, 'ani.mp4'), writer=writervideo)
+        writervideo = animation.PillowWriter(fps=30)
+        anima.save(os.path.join(save_path, 'ani.gif'), writer=writervideo)
+
         plt.show(block=True)
 
 

requirements.txt

@@ -1,8 +1,8 @@
+numpy==1.21.2
 matplotlib==3.4.3
+torch==1.9.0
 torch_geometric==2.0.1
 torch_cluster==1.5.9
-torch==1.9.0
-numpy==1.21.2
 absl_py==0.13.0
 tfrecord==1.14.1
 torch_scatter==2.0.8

run_model.py

@@ -251,7 +251,8 @@ def process_trajectory(trajectory_data, params, model_type, dataset_dir, add_tar
     trajectory = {}
     # decode bytes into corresponding dtypes
     for key, value in trajectory_data.items():
-        raw_data = value.numpy().tobytes()
+        # raw_data = value.numpy().tobytes()
+        raw_data = value[0][0]
         mature_data = np.frombuffer(raw_data, dtype=getattr(np, dtypes[key]))
         mature_data = torch.from_numpy(mature_data).to(device)
         reshaped_data = torch.reshape(mature_data, shapes[key])