more

src/lerobot/policies/pi05_full/annotate/annotate_libero.sh

@@ -16,12 +16,13 @@ TEMPERATURE=0.9
 SAMPLE_INTERVAL=5.0  # generate dialogue every 1 second (all episodes processed)
 
 # Run subtask annotation
-# python /admin/home/jade_choghari/lerobot/src/lerobot/policies/pi05_full/annotate/subtask_annotate.py \
-#     --repo-id "$REPO_ID" \
-#     --video-key observation.images.image \
-#     --output-dir "$OUTPUT_DIR" \
-#     --output-repo-id "jadechoghari/libero-annotate" \
-#     --batch-size "$BATCH_SIZE" \
+python /admin/home/jade_choghari/lerobot/src/lerobot/policies/pi05_full/annotate/subtask_annotate.py \
+    --repo-id "$REPO_ID" \
+    --video-key observation.images.image \
+    --output-dir "$OUTPUT_DIR" \
+    --skip-existing \
+    --output-repo-id "jadechoghari/libero-annotate" \
+    --batch-size "$BATCH_SIZE" \
 # run synthetic data generation (all episodes processed)
 # python examples/dataset/annotate_pgen.py \
 #     --repo-id "$REPO_ID" \
@@ -40,10 +41,10 @@ SAMPLE_INTERVAL=5.0  # generate dialogue every 1 second (all episodes processed)
 # add --push-to-hub flag
 
 # efficient batch processing: 4 episodes at once
-python /admin/home/jade_choghari/lerobot/src/lerobot/policies/pi05_full/annotate/high_level_annotate.py \
-    --repo-id "$REPO_ID" \
-    --output-dir "$OUTPUT_DIR" \
-    --video-mode \
-    --video-key observation.images.image \
-    --video-batch-size "$BATCH_SIZE" \
-    --sample-interval 5.0
+# python /admin/home/jade_choghari/lerobot/src/lerobot/policies/pi05_full/annotate/high_level_annotate.py \
+#     --repo-id "$REPO_ID" \
+#     --output-dir "$OUTPUT_DIR" \
+#     --video-mode \
+#     --video-key observation.images.image \
+#     --video-batch-size "$BATCH_SIZE" \
+#     --sample-interval 5.0

src/lerobot/policies/pi05_full/annotate/subtask_annotate.py

@@ -703,12 +703,19 @@ class SkillAnnotator:
         if skip_existing:
             existing_annotations = load_skill_annotations(dataset.root)
             if existing_annotations and "episodes" in existing_annotations:
-                existing_episode_indices = {int(idx) for idx in existing_annotations["episodes"].keys()}
+                # Only skip episodes that exist AND have non-empty skills
+                existing_episode_indices = set()
+                for idx_str, episode_data in existing_annotations["episodes"].items():
+                    idx = int(idx_str)
+                    # Check if skills list exists and is not empty
+                    if "skills" in episode_data and episode_data["skills"]:
+                        existing_episode_indices.add(idx)
+                
                 original_count = len(episode_indices)
                 episode_indices = [ep for ep in episode_indices if ep not in existing_episode_indices]
                 skipped_count = original_count - len(episode_indices)
                 if skipped_count > 0:
-                    self.console.print(f"[cyan]Skipping {skipped_count} episodes with existing annotations[/cyan]")
+                    self.console.print(f"[cyan]Skipping {skipped_count} episodes with existing non-empty annotations[/cyan]")
 
         if not episode_indices:
             self.console.print("[yellow]No episodes to annotate (all already annotated)[/yellow]")