fix lerobot

benchmarks/openx_by_frame.py

@@ -9,7 +9,7 @@
 import fog_x
 import csv
 import stat
-from fog_x.loader.lerobot import LeRobotLoader
+from fog_x.loader.lerobot import LeRobotLoader_ByFrame
 from fog_x.loader.vla import get_vla_dataloader
 from fog_x.loader.hdf5 import get_hdf5_dataloader
 
@@ -310,7 +310,7 @@ def __init__(
 
     def get_loader(self):
         path = os.path.join(self.exp_dir, "hf")
-        return LeRobotLoader(path, self.dataset_name, batch_size=self.batch_size)
+        return LeRobotLoader_ByFrame(path, self.dataset_name, batch_size=1, slice_length=self.batch_size)
 
     def _recursively_load_data(self, data):
         import torch

evaluation.sh

@@ -3,7 +3,7 @@ sudo echo "Use sudo access for clearning cache"
 
 # Define a list of batch sizes to iterate through
 
-batch_sizes=(4)
+batch_sizes=(64)
 num_batches=200
 # batch_sizes=(1 2)
 
@@ -16,7 +16,7 @@ do
     echo "Running benchmarks with batch size: $batch_size"
 
     # python3 benchmarks/openx.py --dataset_names nyu_door_opening_surprising_effectiveness --num_batches $num_batches --batch_size $batch_size
-    python3 benchmarks/openx_by_frame.py --dataset_names berkeley_cable_routing --num_batches $num_batches --batch_size $batch_size
-    # python3 benchmarks/openx.py --dataset_names bridge --num_batches $num_batches --batch_size $batch_size
+    # python3 benchmarks/openx_by_frame.py --dataset_names berkeley_cable_routing --num_batches $num_batches --batch_size $batch_size
+    python3 benchmarks/openx_by_frame.py --dataset_names bridge --num_batches $num_batches --batch_size $batch_size
     # python3 benchmarks/openx.py --dataset_names berkeley_autolab_ur5 --num_batches $num_batches --batch_size $batch_size
 done

fog_x/loader/lerobot.py

@@ -55,11 +55,11 @@ def get_batch(self):
 
 
 class LeRobotLoader_ByFrame(BaseLoader):
-    def __init__(self, path, dataset_name, batch_size=1, delta_timestamps=None):
-        super(LeRobotLoader, self).__init__(path)
+    def __init__(self, path, dataset_name, batch_size=1, delta_timestamps=None, slice_length=16):
+        super(LeRobotLoader_ByFrame, self).__init__(path)
         self.batch_size = batch_size
         self.dataset = LeRobotDataset(root="/mnt/data/fog_x/hf/", repo_id=dataset_name, delta_timestamps=delta_timestamps)
-        self.episode_index = 0
+        self.slice_length = slice_length
 
     def __len__(self):
         return len(self.dataset.episode_data_index["from"])
@@ -78,33 +78,27 @@ def _frame_to_numpy(frame):
             for attempt in range(max_retries):
                 try:
                     # repeat
-                    if self.episode_index >= len(self.dataset):
-                        self.episode_index = 0
+                    self.episode_index = np.random.randint(0, len(self.dataset))
                     try:
                         from_idx = self.dataset.episode_data_index["from"][self.episode_index].item()
                         to_idx = self.dataset.episode_data_index["to"][self.episode_index].item()
                     except Exception as e:
-                        self.episode_index = 0
                         continue
 
                     # Randomly select random_frames from episode
-                    random_frames = 16
                     episode_length = to_idx - from_idx
-                    if episode_length <= random_frames:
+                    if episode_length <= self.slice_length:
                         random_from = from_idx
                         random_to = to_idx
                     else:
-                        random_from = np.random.randint(from_idx, to_idx - 15)
-                        random_to = random_from + 16
-                    frames = [_frame_to_numpy(self.dataset[idx]) for idx in range(random_from, random_to)]
+                        random_from = np.random.randint(from_idx, to_idx - self.slice_length)
+                        random_to = random_from + self.slice_length
+                    frames = [self.dataset[idx] for idx in range(random_from, random_to)]
                     episode.extend(frames)
-                    self.episode_index += 1
                     break
                 except Exception as e:
                     if attempt == max_retries - 1:
                         raise e
-                    self.episode_index += 1
-
 
             batch_of_episodes.append((episode))