Use scalable annotation runtime, shared human-video module and optional Flora captions

README.md

@@ -174,7 +174,8 @@ features are excluded. Visual questions have a configurable cadence.
 
 A separate **Video generation** category on the Annotations step, next to the text
 annotations. For every subtask (or up to 1–8 per episode), the job turns the robot view into
-a first-person human demonstration, following LeRobot's `--human_video` annotation module:
+a first-person human demonstration, following LeRobot's `--human_video` annotation module
+and the video-generation stages of [Zero-WAM's HumanGen pipeline](https://robbyant-research.github.io/Zero-WAM/):
 
 1. the selected vision model reads the robot segment and writes the human task and an
    image-edit prompt;

frontend/src/main.tsx

@@ -1090,6 +1090,17 @@ function App() {
                       </span>
                     </label>
                   </div>
+                  <p className="field-help">
+                    Follows the video-generation workflow from{" "}
+                    <a
+                      href="https://robbyant-research.github.io/Zero-WAM/"
+                      target="_blank"
+                      rel="noreferrer"
+                    >
+                      Zero-WAM’s HumanGen pipeline
+                    </a>
+                    .
+                  </p>
                   {options.features.includes(HUMAN_VIDEOS) && (
                     <div className="fields-row video-settings">
                       <div>

reporting.py

@@ -53,6 +53,7 @@ def annotation_url(report):
 
 
 def episode_progress(report):
+    """Count successful labeling separately from playable dataset exports."""
     episodes = report.get("episodes", [])
     expected = report["expected_episodes"]
     completed = sum(e["status"] == "completed" for e in episodes)