Update engine to enable openvinotoolkit#3607 and openvinotoolkit#3611

Author: sungchul2

src/otx/engine/engine.py

@@ -367,8 +367,13 @@ def test(
         # NOTE, trainer.test takes only lightning based checkpoint.
         # So, it can't take the OTX1.x checkpoint.
         if checkpoint is not None and not is_ir_ckpt:
+            kwargs_user_input: dict[str, Any] = {}
+            if self.task == OTXTaskType.ZERO_SHOT_VISUAL_PROMPTING:
+                # to update user's custom infer_reference_info_root through cli for zero-shot learning
+                kwargs_user_input.update(infer_reference_info_root=self.model.infer_reference_info_root)
+
             model_cls = model.__class__
-            model = model_cls.load_from_checkpoint(checkpoint_path=checkpoint)
+            model = model_cls.load_from_checkpoint(checkpoint_path=checkpoint, **kwargs_user_input)
 
         if model.label_info != self.datamodule.label_info:
             if (
@@ -462,8 +467,13 @@ def predict(
             datamodule = self._auto_configurator.update_ov_subset_pipeline(datamodule=datamodule, subset="test")
 
         if checkpoint is not None and not is_ir_ckpt:
+            kwargs_user_input: dict[str, Any] = {}
+            if self.task == OTXTaskType.ZERO_SHOT_VISUAL_PROMPTING:
+                # to update user's custom infer_reference_info_root through cli for zero-shot learning
+                kwargs_user_input.update(infer_reference_info_root=self.model.infer_reference_info_root)
+
             model_cls = model.__class__
-            model = model_cls.load_from_checkpoint(checkpoint_path=checkpoint)
+            model = model_cls.load_from_checkpoint(checkpoint_path=checkpoint, **kwargs_user_input)
 
         if model.label_info != self.datamodule.label_info:
             msg = (
@@ -574,8 +584,17 @@ def export(
             )
 
         if not is_ir_ckpt:
+            kwargs_user_input: dict[str, Any] = {}
+            if self.task == OTXTaskType.ZERO_SHOT_VISUAL_PROMPTING:
+                # to update user's custom infer_reference_info_root through cli for zero-shot learning
+                kwargs_user_input.update(infer_reference_info_root=self.model.infer_reference_info_root)
+
             model_cls = self.model.__class__
-            self.model = model_cls.load_from_checkpoint(checkpoint_path=checkpoint, map_location="cpu")
+            self.model = model_cls.load_from_checkpoint(
+                checkpoint_path=checkpoint,
+                map_location="cpu",
+                **kwargs_user_input,
+            )
             self.model.eval()
 
         self.model.explain_mode = explain
@@ -738,8 +757,13 @@ def explain(
             model = self._auto_configurator.get_ov_model(model_name=str(checkpoint), label_info=datamodule.label_info)
 
         if checkpoint is not None and not is_ir_ckpt:
+            kwargs_user_input: dict[str, Any] = {}
+            if self.task == OTXTaskType.ZERO_SHOT_VISUAL_PROMPTING:
+                # to update user's custom infer_reference_info_root through cli for zero-shot learning
+                kwargs_user_input.update(infer_reference_info_root=self.model.infer_reference_info_root)
+
             model_cls = model.__class__
-            model = model_cls.load_from_checkpoint(checkpoint_path=checkpoint)
+            model = model_cls.load_from_checkpoint(checkpoint_path=checkpoint, **kwargs_user_input)
 
         if model.label_info != self.datamodule.label_info:
             msg = (
@@ -841,8 +865,17 @@ def benchmark(
                 )
 
             if not is_ir_ckpt:
+                kwargs_user_input: dict[str, Any] = {}
+                if self.task == OTXTaskType.ZERO_SHOT_VISUAL_PROMPTING:
+                    # to update user's custom infer_reference_info_root through cli for zero-shot learning
+                    kwargs_user_input.update(infer_reference_info_root=self.model.infer_reference_info_root)
+
                 model_cls = self.model.__class__
-                self.model = model_cls.load_from_checkpoint(checkpoint_path=checkpoint, map_location="cpu")
+                self.model = model_cls.load_from_checkpoint(
+                    checkpoint_path=checkpoint,
+                    map_location="cpu",
+                    **kwargs_user_input,
+                )
         elif isinstance(self.model, OVModel):
             msg = "To run benchmark on OV model, checkpoint must be specified."
             raise RuntimeError(msg)