supervisely 6.73.358__py3-none-any.whl → 6.73.360__py3-none-any.whl

supervisely/nn/training/gui/input_selector.py

@@ -12,7 +12,7 @@ from supervisely.project.download import is_cached
 
 class InputSelector:
     title = "Input project"
-    description = "Selected project from which images and annotations will be downloaded"
+    description = "Selected project from which items and annotations will be downloaded"
     lock_message = None
 
     def __init__(self, project_info: ProjectInfo, app_options: dict = {}):

supervisely/nn/training/gui/model_selector.py

@@ -36,6 +36,13 @@ class ModelSelector:
         self.display_widgets = []
         self.app_options = app_options
 
+        model_selector_opts = self.app_options.get("model_selector", {})
+        if not isinstance(model_selector_opts, dict):
+            model_selector_opts = {}
+
+        self.show_pretrained = True
+        self.show_custom = model_selector_opts.get("show_custom", True)
+
         self.team_id = sly_env.team_id()
         self.models = models
 
@@ -51,15 +58,28 @@ class ModelSelector:
                 logger.warning(f"Legacy checkpoints are not available for '{framework}'")
 
         self.experiment_selector = ExperimentSelector(self.team_id, experiment_infos)
+
+        tab_titles = []
+        tab_descriptions = []
+        tab_contents = []
+        if self.show_pretrained:
+            tab_titles.append(ModelSource.PRETRAINED)
+            tab_descriptions.append("Publicly available models")
+            tab_contents.append(self.pretrained_models_table)
+        if self.show_custom:
+            tab_titles.append(ModelSource.CUSTOM)
+            tab_descriptions.append("Models trained by you in Supervisely")
+            tab_contents.append(self.experiment_selector)
+
         self.model_source_tabs = RadioTabs(
-            titles=[ModelSource.PRETRAINED, ModelSource.CUSTOM],
-            descriptions=[
-                "Publicly available models",
-                "Models trained by you in Supervisely",
-            ],
-            contents=[self.pretrained_models_table, self.experiment_selector],
+            titles=tab_titles,
+            descriptions=tab_descriptions,
+            contents=tab_contents,
         )
 
+        if len(tab_titles) > 0:
+            self.model_source_tabs.set_active_tab(tab_titles[0])
+
         self.validator_text = Text("")
         self.validator_text.hide()
         self.button = Button("Select")

supervisely/nn/training/gui/tags_selector.py

@@ -0,0 +1,105 @@
+from supervisely._utils import abs_url, is_debug_with_sly_net, is_development
+from supervisely.app.widgets import Button, Card, Container, TagsTable, Text
+
+
+class TagsSelector:
+    title = "Tags Selector"
+    description = "Select tags that will be used for training"
+    lock_message = "Select training and validation splits to unlock"
+
+    def __init__(self, project_id: int, tags: list, app_options: dict = {}):
+        # Init widgets
+        self.qa_stats_text = None
+        self.tags_table = None
+        self.validator_text = None
+        self.button = None
+        self.container = None
+        self.card = None
+        # -------------------------------- #
+
+        self.display_widgets = []
+        self.app_options = app_options
+
+        # GUI Components
+        if is_development() or is_debug_with_sly_net():
+            qa_stats_link = abs_url(f"projects/{project_id}/stats/datasets")
+        else:
+            qa_stats_link = f"/projects/{project_id}/stats/datasets"
+        self.qa_stats_text = Text(
+            text=f"<i class='zmdi zmdi-chart-donut' style='color: #7f858e'></i> <a href='{qa_stats_link}' target='_blank'> <b> QA & Stats </b></a>"
+        )
+
+        self.tags_table = TagsTable(project_id=project_id)
+        if len(tags) > 0:
+            self.tags_table.select_tags(tags)
+        else:
+            self.tags_table.select_all()
+
+        self.validator_text = Text("")
+        self.validator_text.hide()
+        self.button = Button("Select")
+        self.display_widgets.extend(
+            [self.qa_stats_text, self.tags_table, self.validator_text, self.button]
+        )
+        # -------------------------------- #
+
+        self.container = Container(self.display_widgets)
+        self.card = Card(
+            title=self.title,
+            description=self.description,
+            content=self.container,
+            lock_message=self.lock_message,
+            collapsable=self.app_options.get("collapsable", False),
+        )
+        self.card.lock()
+
+    @property
+    def widgets_to_disable(self) -> list:
+        return [self.tags_table]
+
+    def get_selected_tags(self) -> list:
+        return self.tags_table.get_selected_tags()
+
+    def set_tags(self, tags) -> None:
+        self.tags_table.select_tags(tags)
+
+    def select_all_tags(self) -> None:
+        self.tags_table.select_all()
+
+    def validate_step(self) -> bool:
+        self.validator_text.hide()
+
+        project_tags = self.tags_table.project_meta.tag_metas
+        if len(project_tags) == 0:
+            self.validator_text.set(text="Project has no tags", status="error")
+            self.validator_text.show()
+            return False
+
+        selected_tags = self.tags_table.get_selected_tags()
+        table_data = self.tags_table._table_data
+        empty_tags = [
+            row[0]["data"]
+            for row in table_data
+            if row[0]["data"] in selected_tags and row[2]["data"] == 0 and row[3]["data"] == 0
+        ]
+
+        n_tags = len(selected_tags)
+        if n_tags == 0:
+            message = "Please select at least one tag"
+            status = "error"
+        else:
+            tag_text = "tag" if n_tags == 1 else "tags"
+            message = f"Selected {n_tags} {tag_text}"
+            status = "success"
+            if empty_tags:
+                intersections = set(selected_tags).intersection(empty_tags)
+                if intersections:
+                    tag_text = "tag" if len(intersections) == 1 else "tags"
+                    message += (
+                        f". Selected {tag_text} have no annotations: {', '.join(intersections)}"
+                    )
+                    status = "warning"
+
+        self.validator_text.set(text=message, status=status)
+        self.validator_text.show()
+        return n_tags > 0

supervisely/nn/training/gui/train_val_splits_selector.py

@@ -24,26 +24,51 @@ class TrainValSplitsSelector:
         self.project_id = project_id
 
         # GUI Components
-        self.train_val_splits = TrainValSplits(project_id)
-        train_val_dataset_ids = {"train": [], "val": []}
-        for _, dataset in api.dataset.tree(project_id):
-            if dataset.name.lower() == "train" or dataset.name.lower() == "training":
-                if dataset.items_count > 0:
-                    train_val_dataset_ids["train"].append(dataset.id)
-            elif dataset.name.lower() == "val" or dataset.name.lower() == "validation":
-                if dataset.items_count > 0:
-                    train_val_dataset_ids["val"].append(dataset.id)
-
-        # Check nested dataset names
+        split_methods = self.app_options.get("train_val_split_methods", [])
+        if len(split_methods) == 0:
+            split_methods = ["Random", "Based on tags", "Based on datasets"]
+        random_split = "Random" in split_methods
+        tag_split = "Based on tags" in split_methods
+        ds_split = "Based on datasets" in split_methods
+
+        self.train_val_splits = TrainValSplits(project_id, None, random_split, tag_split, ds_split)
+
+        def _extend_with_nested(root_ds):
+            nested = self.api.dataset.get_nested(self.project_id, root_ds.id)
+            nested_ids = [ds.id for ds in nested]
+            return [root_ds.id] + nested_ids
+
+        train_val_dataset_ids = {"train": set(), "val": set()}
+        for _, dataset in self.api.dataset.tree(self.project_id):
+            ds_name = dataset.name.lower()
+
+            if ds_name in {"train", "training"}:
+                for _id in _extend_with_nested(dataset):
+                    train_val_dataset_ids["train"].add(_id)
+
+            elif ds_name in {"val", "validation", "test", "testing"}:
+                for _id in _extend_with_nested(dataset):
+                    train_val_dataset_ids["val"].add(_id)
+
+        train_val_dataset_ids["train"] = list(train_val_dataset_ids["train"])
+        train_val_dataset_ids["val"] = list(train_val_dataset_ids["val"])
+
         train_count = len(train_val_dataset_ids["train"])
         val_count = len(train_val_dataset_ids["val"])
+
         if train_count > 0 and val_count > 0:
             self.train_val_splits.set_datasets_splits(
                 train_val_dataset_ids["train"], train_val_dataset_ids["val"]
             )
 
         if train_count > 0 and val_count > 0:
-            self.validator_text = Text("Train and val datasets are detected", status="info")
+            if train_count == val_count == 1:
+                self.validator_text = Text("train and val datasets are detected", status="info")
+            else:
+                self.validator_text = Text(
+                    "Multiple train and val datasets are detected. Check manually if selection is correct",
+                    status="info",
+                )
             self.validator_text.show()
         else:
             self.validator_text = Text("")
@@ -71,8 +96,9 @@ class TrainValSplitsSelector:
         split_method = self.train_val_splits.get_split_method()
         warning_text = "Using the same data for training and validation leads to overfitting, poor generalization and biased model selection."
         ensure_text = "Ensure this is intentional."
+        is_valid = False
 
-        if split_method == "Random":
+        def validate_random_split():
             train_ratio = self.train_val_splits.get_train_split_percent()
             val_ratio = self.train_val_splits.get_val_split_percent()
 
@@ -109,8 +135,9 @@ class TrainValSplitsSelector:
                     text="Train and validation splits are selected.",
                     status="success",
                 )
+            return True
 
-        elif split_method == "Based on tags":
+        def validate_based_on_tags():
             train_tag = self.train_val_splits.get_train_tag()
             val_tag = self.train_val_splits.get_val_tag()
 
@@ -130,29 +157,42 @@ class TrainValSplitsSelector:
                 else:
                     tags_count[tag_name] = tag_total
 
-            # @TODO: handle button correctly if validation fails. Do not unlock next card until validation passes if returned False
             if tags_count[train_tag] == 0:
                 self.validator_text.set(
                     text=f"Train tag '{train_tag}' is not present in any images. {ensure_text}",
                     status="error",
                 )
+                return False
             elif tags_count[val_tag] == 0:
                 self.validator_text.set(
                     text=f"Val tag '{val_tag}' is not present in any images. {ensure_text}",
                     status="error",
                 )
-
+                return False
             elif train_tag == val_tag:
                 self.validator_text.set(
                     text=f"Train and val tags are the same. {ensure_text} {warning_text}",
                     status="warning",
                 )
+                return True
             else:
                 self.validator_text.set("Train and val tags are selected", status="success")
+                return True
 
-        elif split_method == "Based on datasets":
+        def validate_based_on_datasets():
             train_dataset_id = self.get_train_dataset_ids()
             val_dataset_id = self.get_val_dataset_ids()
+            if train_dataset_id is None and val_dataset_id is None:
+                self.validator_text.set("No datasets are selected", status="error")
+                return False
+
+            if train_dataset_id is None:
+                self.validator_text.set("No train dataset is selected", status="error")
+                return False
+
+            if val_dataset_id is None:
+                self.validator_text.set("No val dataset is selected", status="error")
+                return False
 
             # Check if datasets are not empty
             stats = self.api.project.get_stats(self.project_id)
@@ -169,6 +209,14 @@ class TrainValSplitsSelector:
                     empty_dataset_names.append(datasets_count[dataset_id]["name"])
 
             if len(empty_dataset_names) > 0:
+                if len(empty_dataset_names) == len(train_dataset_id + val_dataset_id):
+                    empty_ds_text = f"All selected datasets are empty. {ensure_text}"
+                    self.validator_text.set(
+                        text=empty_ds_text,
+                        status="error",
+                    )
+                    return False
+
                 if len(empty_dataset_names) == 1:
                     empty_ds_text = f"Selected dataset: {', '.join(empty_dataset_names)} is empty. {ensure_text}"
                 else:
@@ -178,16 +226,30 @@ class TrainValSplitsSelector:
                     text=empty_ds_text,
                     status="error",
                 )
+                return True
 
             elif train_dataset_id == val_dataset_id:
                 self.validator_text.set(
                     text=f"Same datasets are selected for both train and val splits. {ensure_text} {warning_text}",
                     status="warning",
                 )
+                return True
             else:
                 self.validator_text.set("Train and val datasets are selected", status="success")
+                return True
+
+        if split_method == "Random":
+            is_valid = validate_random_split()
+
+        elif split_method == "Based on tags":
+            is_valid = validate_based_on_tags()
+
+        elif split_method == "Based on datasets":
+            is_valid = validate_based_on_datasets()
+
+        # @TODO: handle button correctly if validation fails. Do not unlock next card until validation passes if returned False
         self.validator_text.show()
-        return True
+        return is_valid
 
     def set_sly_project(self, project: Project) -> None:
         self.train_val_splits._project_fs = project