fix: loadable dataset

We should fix the transform as it generates ugly code in the dataset side.

clip_eval/dataset/encord_dataset.py

@@ -32,9 +32,12 @@ def __getitem__(self, idx):
         img = Image.open(frame_path)
         label = self._labels[idx]
 
-        if self.transform:
-            img, label = self.transform(img, label)
-        return img, label
+        if self.transform is not None:
+            _d = self.transform(dict(image=[img], label=[label]))
+            res_item = dict(image=_d["image"][0], label=_d["label"][0])
+        else:
+            res_item = dict(image=img, label=label)
+        return res_item
 
     def __len__(self):
         return len(self._frame_paths)
@@ -95,6 +98,7 @@ def _setup(
 
         self._frame_paths = []
         self._labels = []
+        class_name_to_idx = {name: idx for idx, name in enumerate(self.class_names)}  # Fast lookup of class indices
         for label_row in self._project.list_label_rows_v2():
             anns_path = self._get_label_row_annotations(label_row)
             label_row.from_labels_dict(json.loads(anns_path.read_text(encoding="utf-8")))
@@ -111,4 +115,4 @@ def _setup(
                     continue
 
                 self._frame_paths.append(self._get_frame_path(label_row, frame_view.frame))
-                self._labels.append(clf_answer.title)
+                self._labels.append(class_name_to_idx[clf_answer.title])