Cache "cache_key" in the pipeline

Signed-off-by: Songki Choi <songki.choi@intel.com>

Author: Songki Choi

src/otx/algorithms/common/adapters/mmcv/pipelines/load_image_from_otx_dataset.py

@@ -45,8 +45,11 @@ def _get_unique_key(results: Dict[str, Any]) -> Tuple:
         # TODO: We should improve it by assigning an unique id to DatasetItemEntity.
         # This is because there is a case which
         # d_item.media.path is None, but d_item.media.data is not None
+        if "cache_key" in results:
+            return results["cache_key"]
         d_item = results["dataset_item"]
-        return d_item.media.path, d_item.roi.id
+        results["cache_key"] = d_item.media.path, d_item.roi.id
+        return results["cache_key"]
 
     def __call__(self, results: Dict[str, Any]):
         """Callback function of LoadImageFromOTXDataset."""
@@ -177,7 +180,6 @@ def _save_cache(self, results: Dict[str, Any]):
             return
         key = self._get_unique_key(results)
         meta = results.copy()
-        meta.pop("dataset_item")  # remove irrlevant info
         img = meta.pop("img")
         self._mem_cache_handler.put(key, img, meta)
 

src/otx/algorithms/detection/adapters/mmdet/datasets/pipelines/load_pipelines.py

@@ -119,7 +119,7 @@ def _load_masks(results, ann_info):
 
     def __call__(self, results: Dict[str, Any]):
         """Callback function of LoadAnnotationFromOTXDataset."""
-        dataset_item = results.get("dataset_item")
+        dataset_item = results.pop("dataset_item")  # Prevent unnecessary deepcopy
         label_list = results.pop("ann_info")["label_list"]
         ann_info = get_annotation_mmdet_format(dataset_item, label_list, self.domain, self.min_size)
         if self.with_bbox: