Enable tracking-only inference pipeline from gui.

If you use "run inference" and don't select any models but do
select a tracking method, it will just re-assign tracks for the
predicted instances you already have in the project.

sleap/gui/inference.py

@@ -877,6 +877,7 @@ def run_learning_pipeline(
         trained_job_paths=trained_job_paths,
         inference_params=inference_params,
         frames_to_predict=frames_to_predict,
+        labels_filename=labels_filename,
     )
 
     return new_labeled_frame_count
@@ -969,6 +970,7 @@ def run_gui_inference(
     trained_job_paths: List[str],
     frames_to_predict: Dict[Video, List[int]],
     inference_params: Dict[str, str],
+    labels_filename: str,
     gui: bool = True,
 ) -> int:
     """Run inference on specified frames using models from training_jobs.
@@ -1015,6 +1017,7 @@ def waiting():
                 trained_job_paths=trained_job_paths,
                 kwargs=inference_params,
                 waiting_callback=waiting,
+                labels_filename=labels_filename,
             )
 
             if success:

sleap/nn/inference.py

@@ -477,9 +477,16 @@ def predict_subprocess(
         kwargs: Dict[str, str],
         frames: Optional[List[int]] = None,
         waiting_callback: Optional[Callable] = None,
+        labels_filename: Optional[str] = None,
     ):
 
-        cli_args = ["python", "-m", "sleap.nn.inference", video.filename]
+        cli_args = ["python", "-m", "sleap.nn.inference"]
+
+        if not trained_job_paths and "tracking.tracker" in kwargs and labels_filename:
+            # No models so we must want to re-track previous predictions
+            cli_args.append(labels_filename)
+        else:
+            cli_args.append(video.filename)
 
         # TODO: better support for video params
         if hasattr(video.backend, "dataset"):