Bug/sg 000 merge failure for datasetparams (#1140)

* failure verification

* added set_struct before merge

* added set_struct before merge in val too

* release tag filter added back

* return type fixed

* used merge_with with open_dict

* added test

src/super_gradients/training/dataloaders/dataloaders.py

@@ -3,7 +3,7 @@
 import hydra
 import numpy as np
 import torch
-from omegaconf import OmegaConf, UnsupportedValueType
+from omegaconf import OmegaConf, UnsupportedValueType, DictConfig, open_dict
 from torch.utils.data import BatchSampler, DataLoader, TensorDataset, RandomSampler
 
 import super_gradients
@@ -101,12 +101,19 @@ def _process_dataset_params(cfg, dataset_params, train: bool):
         # >>> dataset_params = OmegaConf.merge(default_dataset_params, dataset_params)
         # >>> return hydra.utils.instantiate(dataset_params)
         # For some reason this breaks interpolation :shrug:
-
+        if not isinstance(dataset_params, DictConfig):
+            dataset_params = OmegaConf.create(dataset_params)
         if train:
-            cfg.train_dataset_params = OmegaConf.merge(cfg.train_dataset_params, dataset_params)
+            train_dataset_params = cfg.train_dataset_params
+            with open_dict(train_dataset_params):
+                train_dataset_params.merge_with(dataset_params)
+            cfg.train_dataset_params = train_dataset_params
             return hydra.utils.instantiate(cfg.train_dataset_params)
         else:
-            cfg.val_dataset_params = OmegaConf.merge(cfg.val_dataset_params, dataset_params)
+            val_dataset_params = cfg.val_dataset_params
+            with open_dict(val_dataset_params):
+                val_dataset_params.merge_with(dataset_params)
+            cfg.val_dataset_params = val_dataset_params
             return hydra.utils.instantiate(cfg.val_dataset_params)
 
     except UnsupportedValueType:

tests/unit_tests/detection_dataset_test.py

@@ -1,13 +1,31 @@
 import unittest
 from pathlib import Path
+from typing import Dict
 
-from super_gradients.training.dataloaders import coco2017_train_yolo_nas
+from torch.utils.data import DataLoader
+from super_gradients.training.dataloaders import coco2017_train_yolo_nas, get_data_loader
 from super_gradients.training.datasets import COCODetectionDataset
 from super_gradients.training.datasets.data_formats.default_formats import LABEL_CXCYWH
 from super_gradients.training.exceptions.dataset_exceptions import DatasetValidationException, ParameterMismatchException
 from super_gradients.training.transforms import DetectionMosaic, DetectionTargetsFormatTransform, DetectionPaddedRescale
 
 
+class DummyCOCODetectionDatasetInheritor(COCODetectionDataset):
+    def __init__(self, json_file: str, subdir: str, dummy_field: int, *args, **kwargs):
+        super(DummyCOCODetectionDatasetInheritor, self).__init__(json_file=json_file, subdir=subdir, *args, **kwargs)
+        self.dummy_field = dummy_field
+
+
+def dummy_coco2017_inheritor_train_yolo_nas(dataset_params: Dict = None, dataloader_params: Dict = None) -> DataLoader:
+    return get_data_loader(
+        config_name="coco_detection_yolo_nas_dataset_params",
+        dataset_cls=DummyCOCODetectionDatasetInheritor,
+        train=True,
+        dataset_params=dataset_params,
+        dataloader_params=dataloader_params,
+    )
+
+
 class DetectionDatasetTest(unittest.TestCase):
     def setUp(self) -> None:
         self.mini_coco_data_dir = str(Path(__file__).parent.parent / "data" / "tinycoco")
@@ -23,7 +41,6 @@ def test_normal_coco_dataset_creation(self):
         COCODetectionDataset(**train_dataset_params)
 
     def test_coco_dataset_creation_with_wrong_classes(self):
-
         train_dataset_params = {
             "data_dir": self.mini_coco_data_dir,
             "subdir": "images/train2017",
@@ -88,6 +105,25 @@ def test_coco_detection_dataset_override_with_objects(self):
         self.assertEqual(batch[0].shape[2], 384)
         self.assertEqual(batch[0].shape[3], 384)
 
+    def test_coco_detection_dataset_override_with_new_entries(self):
+        train_dataset_params = {
+            "data_dir": self.mini_coco_data_dir,
+            "input_dim": 384,
+            "transforms": [
+                DetectionMosaic(input_dim=384),
+                DetectionPaddedRescale(input_dim=384, max_targets=10),
+                DetectionTargetsFormatTransform(max_targets=10, output_format=LABEL_CXCYWH),
+            ],
+            "dummy_field": 10,
+        }
+        train_dataloader_params = {"num_workers": 0}
+        dataloader = dummy_coco2017_inheritor_train_yolo_nas(dataset_params=train_dataset_params, dataloader_params=train_dataloader_params)
+        batch = next(iter(dataloader))
+        print(batch[0].shape)
+        self.assertEqual(batch[0].shape[2], 384)
+        self.assertEqual(batch[0].shape[3], 384)
+        self.assertEqual(dataloader.dataset.dummy_field, 10)
+
 
 if __name__ == "__main__":
     unittest.main()