diff --git a/src/nrtk_explorer/app/core.py b/src/nrtk_explorer/app/core.py
index 952c7323..1e39f58d 100644
--- a/src/nrtk_explorer/app/core.py
+++ b/src/nrtk_explorer/app/core.py
@@ -23,6 +23,12 @@
     config_features_to_enabled_features,
     config_preset_to_enabled_features,
 )
+from nrtk_explorer.app.images.image_ids import (
+    GROUND_TRUTH_MODEL,
+)
+from nrtk_explorer.app.images.stateful_annotations import (
+    make_stateful_annotations,
+)
 from nrtk_explorer.app.images.images import Images
 from nrtk_explorer.app.images.image_server import ImageServer
 from nrtk_explorer.app.applet import Applet
@@ -132,6 +138,8 @@ def __init__(self, server=None, **kwargs):
 
         images = Images(server=self.server)
         self._image_server = ImageServer(server=self.server, images=images)
+        ground_truth_annotations = make_stateful_annotations(self.server, GROUND_TRUTH_MODEL)
+        self.state.inference_models_obj = {0: {"name": "groundtruth"}}
 
         self._datasets_app = None
         if self.datasets_enabled:
@@ -150,15 +158,23 @@ def __init__(self, server=None, **kwargs):
             from nrtk_explorer.app.features.transforms import TransformsApp
 
             self._transforms_app = TransformsApp(
-                server=self.server.create_child_server(), images=images, **kwargs
+                server=self.server.create_child_server(),
+                images=images,
+                ground_truth_annotations=ground_truth_annotations,
+                **kwargs,
             )
 
+            self.context.tramsforms_app = self._transforms_app
+
         self._images_app = None
         if self.images_enabled:
             from nrtk_explorer.app.features.images import ImagesApp
 
             self._images_app = ImagesApp(
-                server=self.server.create_child_server(), images=images, **kwargs
+                server=self.server.create_child_server(),
+                images=images,
+                ground_truth_annotations=ground_truth_annotations,
+                **kwargs,
             )
 
         self._inference_app = None
@@ -207,6 +223,7 @@ def __init__(self, server=None, **kwargs):
         self.state.dataset_ids = []
         self.state.hovered_id = None
         self.state.maximised_id = None
+        self.state.maximised_annotations = []
 
         def clear_hovered(**kwargs):
             self.state.hovered_id = None
diff --git a/src/nrtk_explorer/app/features/images.py b/src/nrtk_explorer/app/features/images.py
index 14476e50..183a1aaf 100644
--- a/src/nrtk_explorer/app/features/images.py
+++ b/src/nrtk_explorer/app/features/images.py
@@ -83,7 +83,7 @@ def __init__(
 
         self.images = images or Images(server)
 
-        ground_truth_annotations = ground_truth_annotations or make_stateful_annotations(
+        self.ground_truth_annotations = ground_truth_annotations or make_stateful_annotations(
             server, GROUND_TRUTH_MODEL
         )
         self.context.ground_truth_annotations = ground_truth_annotations.annotations_factory
diff --git a/src/nrtk_explorer/app/features/inference.py b/src/nrtk_explorer/app/features/inference.py
index 8ba4c6c3..0bfa1fbd 100644
--- a/src/nrtk_explorer/app/features/inference.py
+++ b/src/nrtk_explorer/app/features/inference.py
@@ -46,7 +46,7 @@ def __init__(
         self.state.inference_models = [self.state.inference_models_options[0]]
         self.state.inference_multi_model = False
 
-        inference_models_obj = {0: {"name": "ground-truth"}}
+        inference_models_obj = {0: {"name": "groundtruth"}}
         for i, model in enumerate(self.state.inference_models):
             inference_models_obj[i + 1] = {"name": model}
 
@@ -101,7 +101,7 @@ def update_inference_models(self, models):
                     "transformed_annotations": transformed_annotations.annotations_factory,
                 }
 
-        models_obj = {0: {"name": "ground-truth"}}
+        models_obj = {0: {"name": "groundtruth"}}
         for i, model in enumerate(models):
             models_obj[i + 1] = {"name": model}
 
diff --git a/src/nrtk_explorer/app/features/parameters.py b/src/nrtk_explorer/app/features/parameters.py
index 3e666b43..01501e3b 100644
--- a/src/nrtk_explorer/app/features/parameters.py
+++ b/src/nrtk_explorer/app/features/parameters.py
@@ -115,6 +115,15 @@ def transform_apply_ui(self):
                 flat=True,
             )
 
+    def transform_preview_ui(self):
+        with html.Div(trame_server=self.server):
+            quasar.QBtn(
+                "Preview Transforms",
+                click=(self.server.controller.preview_transform),
+                classes="full-width",
+                flat=True,
+            )
+
     def transforms_ui(self):
         with html.Div(trame_server=self.server):
             TransformsWidget(
diff --git a/src/nrtk_explorer/app/features/preview.py b/src/nrtk_explorer/app/features/preview.py
new file mode 100644
index 00000000..ca9f360f
--- /dev/null
+++ b/src/nrtk_explorer/app/features/preview.py
@@ -0,0 +1,328 @@
+from typing import Dict
+
+from trame.ui.quasar import QLayout
+from trame.widgets import quasar
+from trame.widgets import html
+from trame.app import get_server
+from trame_client.widgets.trame import Getter
+
+from trame_annotations.widgets.annotations import ImageDetection
+from nrtk_explorer.app.images.image_server import ImageServer
+from nrtk_explorer.widgets.nrtk_explorer import AnnotationAggregator
+
+from nrtk_explorer.library.app_config import process_config
+from nrtk_explorer.library.dataset import (
+    get_dataset,
+    expand_hugging_face_datasets,
+    dataset_select_options,
+)
+
+import nrtk_explorer.library.transforms as trans
+import nrtk_explorer.library.yaml_transforms as nrtk_yaml
+import nrtk_explorer.library.serialization_helpers as serialization_helpers
+
+from nrtk_explorer.app.applet import Applet
+from nrtk_explorer.app.features.parameters import ParametersApp
+from nrtk_explorer.app.images.images import Images
+
+from nrtk_explorer.app.images.stateful_annotations import (
+    make_stateful_annotations,
+)
+
+from nrtk_explorer.app.images.image_ids import (
+    GROUND_TRUTH_MODEL,
+)
+
+
+class PreviewApp(Applet):
+    def __init__(
+        self,
+        server,
+        images=None,
+        ground_truth_annotations=None,
+        **kwargs,
+    ):
+        super().__init__(server)
+
+        self.images = images or Images(server)
+
+        self._image_server = ImageServer(server=self.server, images=self.images)
+
+        self.ground_truth_annotations = ground_truth_annotations or make_stateful_annotations(
+            server, GROUND_TRUTH_MODEL
+        )
+        self.context.ground_truth_annotations = self.ground_truth_annotations.annotations_factory
+
+        if self.context.parameters_app is None:
+            self.context.parameters_app = ParametersApp(
+                server=server,
+            )
+
+        self._parameters_app = self.context.parameters_app
+
+        self.state.dataset_ids = []
+        self.state.preview_inference_models_obj = {0: {"name": "groundtruth"}}
+        self.state.preview_image_id = None
+        self.state.transformed_preview_image = None
+
+        self._ui = None
+
+        self._transform_classes: Dict[str, type[trans.ImageTransform]] = {
+            "blur": trans.GaussianBlurTransform,
+            "invert": trans.InvertTransform,
+            "downsample": trans.DownSampleTransform,
+            "identity": trans.IdentityTransform,
+        }
+
+        # Add transform from YAML definition
+        self._transform_classes.update(nrtk_yaml.generate_transforms())
+
+        self._parameters_app._transform_classes = self._transform_classes
+
+        # Initialize the transforms pipeline to the identity
+        self._parameters_app._default_transform = "blur"
+        self._parameters_app.on_add_transform()
+
+        self.server.controller.on_server_ready.add(self.on_server_ready)
+
+    def on_server_ready(self, *args, **kwargs):
+        self.server.controller.preview_transform.add(self.on_preview_transform)
+        self.state.change("preview_image_id")(self.on_preview_transform)
+        self.on_preview_transform()
+
+    def on_preview_transform(self, **kwargs):
+        preview_id = self.state.preview_image_id
+        if preview_id is None:
+            return
+
+        transforms = list(map(lambda t: t["instance"], self.context.transforms))
+
+        for transform in transforms:
+            params = transform.get_parameters()
+            for key, value in transform.get_parameters_description().items():
+                if "deserialize_func" in value.keys():
+                    params[key] = getattr(serialization_helpers, value["deserialize_func"])(
+                        params[key]
+                    )
+            transform.set_parameters(params)
+
+        chained_transform = trans.ChainedImageTransform(transforms)
+
+        image = self.images.get_image_without_cache_eviction(preview_id)
+        transformed_image = chained_transform.execute(image)
+        import base64
+        from io import BytesIO
+
+        buffered = BytesIO()
+        transformed_image.save(buffered, format="JPEG")
+        img_str = base64.b64encode(buffered.getvalue())
+        img_base64 = bytes("data:image/jpeg;base64,", encoding="utf-8") + img_str
+        img_base64_str = img_base64.decode("utf-8")
+        self.state.transformed_preview_image = img_base64_str
+
+    def settings_widget(self):
+        with html.Div(classes="col"):
+            self._parameters_app.transforms_ui()
+
+    def preview_ui(self):
+        with html.Div():
+            self._parameters_app.transform_preview_ui()
+
+    def carousel_ui(self):
+        with quasar.QCarousel(
+            v_model=("preview_image_id",),
+            style="padding-bottom: 6rem;",
+            classes="fit",
+            control_color="primary",
+            animated=True,
+            infinite=True,
+            swipeable=True,
+            navigation=True,
+            thumbnails=True,
+            arrows=True,
+            padding=True,
+            transition_prev="slide-right",
+            transition_next="slide-left",
+        ):
+            with html.Template(
+                raw_attrs=['v-slot:navigation-icon="{ index, name, active, btnProps, onClick }"']
+            ):
+                with Getter(name=("`img_${name}`",)):
+                    quasar.QImg(
+                        key=("name",),
+                        classes="rounded-borders q-mr-md",
+                        style=(
+                            "active ? 'width: 6rem; height: 6rem; border-style: solid; border-width: 0.125rem; border-color: red;' : 'width: 6rem; height: 6rem;'",
+                        ),
+                        fit="cover",
+                        src=("value",),
+                        click="onClick",
+                    )
+
+            with html.Template(v_for=("identifier in dataset_ids",)):
+                with quasar.QCarouselSlide(
+                    name=("identifier",),
+                    key=("identifier",),
+                ):
+                    with html.Div(
+                        classes="row fit justify-start items-center q-gutter-xs q-col-gutter no-wrap",
+                    ):
+                        with AnnotationAggregator(
+                            image_id=("identifier",),
+                            transformed=False,
+                            models=("preview_inference_models_obj",),
+                        ):
+                            with Getter(name=("`img_${identifier}`",)):
+                                with html.Div(
+                                    classes="rounded-borders col-6 full-height",
+                                ):
+                                    ImageDetection(
+                                        identifier=("identifier",),
+                                        src=("value",),
+                                        models=("preview_inference_models_obj",),
+                                        annotations=("aggregateAnnotations",),
+                                        categories=("annotation_categories",),
+                                        container_selector=".row",
+                                        score_threshold=("confidence_score_threshold", 0.8),
+                                        color_by="model",
+                                    )
+
+                        with AnnotationAggregator(
+                            image_id=("identifier",),
+                            transformed=True,
+                            models=("preview_inference_models_obj",),
+                        ):
+                            with html.Div(
+                                classes="rounded-borders col-6 full-height",
+                            ):
+                                ImageDetection(
+                                    identifier=("identifier",),
+                                    src=("transformed_preview_image",),
+                                    models=("preview_inference_models_obj",),
+                                    # annotations=("groundtruth_annotations",),
+                                    annotations=("aggregateAnnotations",),
+                                    categories=("annotation_categories",),
+                                    container_selector=".row",
+                                    score_threshold=("confidence_score_threshold", 0.8),
+                                    color_by="model",
+                                )
+
+    # This is only used within when this module (file) is executed as an Standalone app.
+    @property
+    def ui(self):
+        if self._ui is None:
+            with QLayout(
+                self.server, view="lhh LpR lff", classes="shadow-2 rounded-borders bg-grey-2"
+            ) as layout:
+                # # Toolbar
+                with quasar.QHeader():
+                    with quasar.QToolbar(classes="shadow-4"):
+                        quasar.QBtn(
+                            flat=True,
+                            click="drawerLeft = !drawerLeft",
+                            round=True,
+                            dense=False,
+                            icon="menu",
+                        )
+                        quasar.QToolbarTitle("Transforms")
+
+                # # Main content
+                with quasar.QPageContainer():
+                    with quasar.QPage(classes="row"):
+                        with html.Div(classes="col-2 q-pa-md"):
+                            self.settings_widget()
+                            self.preview_ui()
+
+                        with html.Div(classes="col-10 q-pa-md bg-grey-3 shadow-2 rounded-borders"):
+                            self.carousel_ui()
+
+                self._ui = layout
+        return self._ui
+
+
+def load_dataset(server, **kwargs):
+    import os
+    import nrtk_explorer.test_data
+
+    DIR_NAME = os.path.dirname(nrtk_explorer.test_data.__file__)
+    DEFAULT_DATASETS = [
+        f"{DIR_NAME}/coco-od-2017/test_val2017.json",
+    ]
+    NUM_IMAGES_DEFAULT = 500
+
+    config_options = {
+        "dataset": {
+            "flags": ["--dataset"],
+            "params": {
+                "nargs": "+",
+                "default": DEFAULT_DATASETS,
+                "help": "Path to the JSON file describing the image dataset",
+            },
+        },
+    }
+    config = process_config(server.cli, config_options, **kwargs)
+
+    server.state.input_datasets = expand_hugging_face_datasets(config["dataset"])
+
+    server.state.all_datasets = server.state.input_datasets
+    server.state.all_datasets_options = dataset_select_options(server.state.all_datasets)
+
+    with server.state:
+        server.state.num_images = NUM_IMAGES_DEFAULT
+        server.state.num_images_max = 0
+        server.state.dataset_ids = []  # sampled images
+        server.state.user_selected_ids = []  # ensure image update in transforms app via image list
+        server.context.dataset = get_dataset(server.state.all_datasets[0])
+        server.state.num_images_max = len(server.context.dataset.imgs)
+        server.state.num_images = min(server.state.num_images_max, server.state.num_images)
+        server.state.dirty("num_images")  # Trigger resample_images()
+        server.state.random_sampling_disabled = False
+        server.state.num_images_disabled = False
+
+        server.state.annotation_categories = {
+            category["id"]: category for category in server.context.dataset.cats.values()
+        }
+
+
+def resample_images(server, images, **kwargs):
+    import random
+
+    ids = [image["id"] for image in server.context.dataset.imgs.values()]
+
+    selected_images = []
+    if server.state.num_images:
+        if server.state.random_sampling:
+            selected_images = random.sample(ids, min(len(ids), server.state.num_images))
+        else:
+            selected_images = ids[: server.state.num_images]
+    else:
+        selected_images = ids
+
+    with server.state:
+        server.context.dataset_ids = selected_images
+        server.state.dataset_ids = [str(id) for id in server.context.dataset_ids]
+        server.state.user_selected_ids = server.state.dataset_ids
+        if len(server.state.dataset_ids) > 0:
+            server.state.preview_image_id = server.state.dataset_ids[0]
+        else:
+            server.state.preview_image_id = None
+
+        server.context.ground_truth_annotations.get_annotations(server.state.dataset_ids)
+
+
+def main(server=None, *args, **kwargs):
+    server = get_server(client_type="vue3")
+    images = Images(server)
+
+    load_dataset(server)
+
+    transforms_app = PreviewApp(server, images)
+    transforms_app.ui
+
+    resample_images(server, images)
+
+    server.start(**kwargs)
+
+
+if __name__ == "__main__":
+    main()
diff --git a/src/nrtk_explorer/app/features/transforms.py b/src/nrtk_explorer/app/features/transforms.py
index 57ce4722..141164bd 100644
--- a/src/nrtk_explorer/app/features/transforms.py
+++ b/src/nrtk_explorer/app/features/transforms.py
@@ -12,6 +12,7 @@
 from nrtk_explorer.app.applet import Applet
 from nrtk_explorer.app.trame_utils import ProcessingStep
 from nrtk_explorer.app.features.parameters import ParametersApp
+from nrtk_explorer.app.features.preview import PreviewApp
 from nrtk_explorer.app.images.images import Images
 
 from nrtk_explorer.app.ui.image_list import (
@@ -36,6 +37,10 @@ def __init__(
             server=server,
         )
 
+        self.context.parameters_app = self._parameters_app
+
+        self._preview_app = PreviewApp(self.server, self.images, **kwargs)
+
         self._ui = None
 
         self._transform_classes: Dict[str, type[trans.ImageTransform]] = {
@@ -66,6 +71,7 @@ def __init__(
         )
 
         self.server.controller.apply_transform.add(self.on_apply_transform)
+        self.server.controller.preview_transform.add(self.on_preview_transform)
         self.server.controller.on_server_ready.add(self.on_server_ready)
 
     def on_server_ready(self, *args, **kwargs):
@@ -73,6 +79,7 @@ def on_server_ready(self, *args, **kwargs):
 
     def on_apply_transform(self, **kwargs):
         # Turn on switch if user clicked lower apply button
+        self.state.show_preview = False
         self.state.transform_enabled_switch = True
         transforms = list(map(lambda t: t["instance"], self.context.transforms))
 
@@ -90,14 +97,45 @@ def on_apply_transform(self, **kwargs):
         if self.ctrl.run_transform.exists():
             self.ctrl.run_transform()
 
+    def on_preview_transform(self, **kwargs):
+        if self.state.preview_image_id is None and len(self.state.dataset_ids) > 0:
+            self.state.preview_image_id = self.state.dataset_ids[0]
+
+        self.state.show_preview = True
+
     def settings_widget(self):
         with html.Div(classes="col"):
             self._parameters_app.transforms_ui()
 
     def apply_ui(self):
         with html.Div():
+            self._parameters_app.transform_preview_ui()
             self._parameters_app.transform_apply_ui()
 
+        with quasar.QDialog(
+            full_width=True,
+            full_height=True,
+            transition_duration=0,
+            v_model=("show_preview", False),
+        ):
+            with html.Div(classes="row bg-grey-3 shadow-2 rounded-borders"):
+                quasar.QBtn(
+                    icon="close",
+                    flat=True,
+                    round=True,
+                    dense=True,
+                    click="show_preview = false;",
+                    style="position: absolute; margin: 0.5rem;",
+                )
+                with html.Div(classes="col-2 q-pa-md"):
+                    html.H5("Transform Preview")
+                    self._parameters_app.transforms_ui()
+                    self._parameters_app.transform_preview_ui()
+                    self._parameters_app.transform_apply_ui()
+
+                with html.Div(classes="col-10 q-pa-md"):
+                    self._preview_app.carousel_ui()
+
     # This is only used within when this module (file) is executed as an Standalone app.
     @property
     def ui(self):
diff --git a/src/nrtk_explorer/app/images/image_ids.py b/src/nrtk_explorer/app/images/image_ids.py
index 2a1efa4a..a71ec3ac 100644
--- a/src/nrtk_explorer/app/images/image_ids.py
+++ b/src/nrtk_explorer/app/images/image_ids.py
@@ -1,6 +1,6 @@
 from nrtk_explorer.app.images.image_meta import dataset_id_to_meta
 
-GROUND_TRUTH_MODEL = "ground-truth"
+GROUND_TRUTH_MODEL = "groundtruth"
 
 
 def image_id_to_dataset_id(image_id: str):
@@ -15,6 +15,10 @@ def dataset_id_to_transformed_image_id(dataset_id: str):
     return f"transformed_img_{dataset_id}"
 
 
+def dataset_id_to_preview_image_id(dataset_id: str):
+    return f"preview_img_{dataset_id}"
+
+
 def image_id_to_result_id(image_id: str, model_name: str):
     return f"result_{image_id}_{model_name}"
 
@@ -27,6 +31,10 @@ def is_transformed(image_id: str):
     return image_id.startswith("transformed_img_")
 
 
+def is_preview(image_id: str):
+    return image_id.startswith("preview_img_")
+
+
 def get_image_state_keys(dataset_id: str, annotation_models: list[str]):
     image_id = dataset_id_to_image_id(dataset_id)
     transformed_image_id = dataset_id_to_transformed_image_id(dataset_id)
diff --git a/src/nrtk_explorer/app/ui/image_list.py b/src/nrtk_explorer/app/ui/image_list.py
index 69809440..a49f52a9 100644
--- a/src/nrtk_explorer/app/ui/image_list.py
+++ b/src/nrtk_explorer/app/ui/image_list.py
@@ -288,7 +288,7 @@ def __init__(self, **kwargs):
                                 }, {});
 
                                 const transformedAnnotations = Object.entries(inference_models_obj).reduce(function(acc, [model_id, model]){
-                                    if (model.name == 'ground-truth') {
+                                    if (model.name == 'groundtruth') {
                                         acc[model_id] = get(`result_${original_id}_${model.name}`);
                                     } else {
                                         acc[model_id] = get(`result_${transformed_id}_${model.name}`);
@@ -312,7 +312,7 @@ def __init__(self, **kwargs):
                                     original_src: get(original_id).value,
                                     transformed: transformed_id,
                                     transformed_src: get(transformed_id).value,
-                                    groundTruthAnnotations: get(`result_${original_id}_ground-truth`) || [],
+                                    groundTruthAnnotations: get(`result_${original_id}_groundtruth`) || [],
                                     originalAnnotations: originalAnnotations || {},
                                     transformedAnnotations: transformedAnnotations,
                                     originalScores,
diff --git a/src/nrtk_explorer/library/embeddings_extractor.py b/src/nrtk_explorer/library/embeddings_extractor.py
index 7f4f75fb..a7e6dfcd 100644
--- a/src/nrtk_explorer/library/embeddings_extractor.py
+++ b/src/nrtk_explorer/library/embeddings_extractor.py
@@ -25,7 +25,13 @@ def __getitem__(self, i):
 
 class EmbeddingsExtractor:
     def __init__(self, model_name="resnet50d", force_cpu=False):
-        self.device = "cuda" if torch.cuda.is_available() and not force_cpu else "cpu"
+        if force_cpu:
+            self.device = "cpu"
+        elif torch.cuda.is_available():
+            self.device = "cuda"
+        else:
+            self.device = "cpu"
+
         self.model = model_name
         self.reset()
 
diff --git a/src/nrtk_explorer/library/predictor.py b/src/nrtk_explorer/library/predictor.py
index 032ca2c7..329eb14e 100644
--- a/src/nrtk_explorer/library/predictor.py
+++ b/src/nrtk_explorer/library/predictor.py
@@ -27,7 +27,14 @@ def __init__(
         force_cpu: bool = False,
     ):
         self.task = task
-        self.device = "cuda" if torch.cuda.is_available() and not force_cpu else "cpu"
+
+        if force_cpu:
+            self.device = "cpu"
+        elif torch.cuda.is_available():
+            self.device = "cuda"
+        else:
+            self.device = "cpu"
+
         self.pipeline = model_name
         self.reset()
 
diff --git a/src/nrtk_explorer/widgets/nrtk_explorer.py b/src/nrtk_explorer/widgets/nrtk_explorer.py
index ff36edeb..ff0325d1 100644
--- a/src/nrtk_explorer/widgets/nrtk_explorer.py
+++ b/src/nrtk_explorer/widgets/nrtk_explorer.py
@@ -119,3 +119,22 @@ def __init__(self, **kwargs):
             "models",
         ]
         self._event_names += []
+
+
+class AnnotationAggregator(HtmlElement):
+    def __init__(self, **kwargs):
+        super().__init__(
+            "annotation-aggregator",
+            **kwargs,
+        )
+        self._attr_names += [
+            ("image_id", "imageId"),
+            "transformed",
+            "models",
+        ]
+        self._event_names += []
+
+        slot_props = [
+            "aggregateAnnotations",
+        ]
+        self._attributes["slot"] = f'v-slot="{{ {", ".join(slot_props)} }}"'
diff --git a/vue-components/src/components/AnnotationAggregator.vue b/vue-components/src/components/AnnotationAggregator.vue
new file mode 100644
index 00000000..2107f6d0
--- /dev/null
+++ b/vue-components/src/components/AnnotationAggregator.vue
@@ -0,0 +1,54 @@
+<script setup lang="ts">
+import { ref, inject, watch, onBeforeMount, onBeforeUnmount } from 'vue'
+
+interface Props {
+  imageId: string
+  transformed: boolean
+  models: { [modelId: string]: any }
+}
+
+const props = defineProps<Props>()
+
+const trame: any = inject('trame')
+
+const aggregateAnnotations = ref({})
+
+function updateAnnotations() {
+  if (props.imageId == undefined || props.models == undefined) {
+    aggregateAnnotations.value = {}
+    return
+  }
+
+  const agg = Object.entries(props.models).reduce(function (acc, [model_id, model]) {
+    const originalId = `img_${props.imageId}`
+    const transformedId = `transformed_img_${props.imageId}`
+
+    let imageId: string = props.transformed ? transformedId : originalId
+
+    if (model.name == 'groundtruth') {
+      acc[model_id] = trame.state.get(`result_${originalId}_${model.name}`)
+    } else {
+      acc[model_id] = trame.state.get(`result_${imageId}_${model.name}`)
+    }
+
+    return acc
+  }, {} as any)
+
+  aggregateAnnotations.value = agg
+}
+
+watch(() => [props.imageId, props.transformed, props.models], updateAnnotations)
+
+onBeforeMount(() => {
+  trame.state.addListener(updateAnnotations)
+  updateAnnotations()
+})
+
+onBeforeUnmount(() => {
+  trame.state.removeListener(updateAnnotations)
+})
+</script>
+
+<template>
+  <slot :aggregateAnnotations></slot>
+</template>
diff --git a/vue-components/src/components/index.js b/vue-components/src/components/index.js
index a78abfcb..ffb947e4 100644
--- a/vue-components/src/components/index.js
+++ b/vue-components/src/components/index.js
@@ -5,6 +5,7 @@ import FilterOptionsWidget from './FilterOptionsWidget.vue'
 import FilterOperatorWidget from './FilterOperatorWidget.vue'
 import ExportWidget from './ExportWidget.vue'
 import ScoreTable from './ScoreTable.vue'
+import AnnotationAggregator from './AnnotationAggregator.vue'
 
 export default {
   scatterPlot: ScatterPlot,
@@ -13,5 +14,6 @@ export default {
   filterOptionsWidget: FilterOptionsWidget,
   filterOperatorWidget: FilterOperatorWidget,
   exportWidget: ExportWidget,
-  scoreTable: ScoreTable
+  scoreTable: ScoreTable,
+  annotationAggregator: AnnotationAggregator
 }