supervisely 6.73.390__py3-none-any.whl → 6.73.392__py3-none-any.whl

supervisely/nn/training/gui/classes_selector.py

@@ -1,14 +1,36 @@
+from typing import List, Tuple
+
 from supervisely._utils import abs_url, is_debug_with_sly_net, is_development
-from supervisely.app.widgets import Button, Card, ClassesTable, Container, Text
+from supervisely.app.widgets import (
+    Button,
+    Card,
+    CheckboxField,
+    ClassesTable,
+    Container,
+    Text,
+)
+from supervisely.geometry.bitmap import Bitmap
+from supervisely.geometry.graph import GraphNodes
+from supervisely.geometry.polygon import Polygon
+from supervisely.geometry.rectangle import Rectangle
+from supervisely.nn.task_type import TaskType
+from supervisely.nn.training.gui.model_selector import ModelSelector
 
 
 class ClassesSelector:
     title = "Classes Selector"
     description = "Select classes that will be used for training"
-    lock_message = "Select training and validation splits to unlock"
+    lock_message = "Select previous step to unlock"
 
-    def __init__(self, project_id: int, classes: list, app_options: dict = {}):
+    def __init__(
+        self,
+        project_id: int,
+        classes: list,
+        model_selector: ModelSelector = None,
+        app_options: dict = {},
+    ):
         # Init widgets
+        self.model_selector = model_selector
         self.qa_stats_text = None
         self.classes_table = None
         self.validator_text = None
@@ -37,9 +59,22 @@ class ClassesSelector:
 
         self.validator_text = Text("")
         self.validator_text.hide()
+
+        self.convert_class_shapes_checkbox = CheckboxField(
+            title="Convert classes to model task type",
+            description="If enabled, classes with compatible shapes will be converted according to the model CV task type",
+            checked=False,
+        )
+
         self.button = Button("Select")
         self.display_widgets.extend(
-            [self.qa_stats_text, self.classes_table, self.validator_text, self.button]
+            [
+                self.qa_stats_text,
+                self.classes_table,
+                self.convert_class_shapes_checkbox,
+                self.validator_text,
+                self.button,
+            ]
         )
         # -------------------------------- #
 
@@ -55,7 +90,7 @@ class ClassesSelector:
 
     @property
     def widgets_to_disable(self) -> list:
-        return [self.classes_table]
+        return [self.classes_table, self.convert_class_shapes_checkbox]
 
     def get_selected_classes(self) -> list:
         return self.classes_table.get_selected_classes()
@@ -66,40 +101,224 @@ class ClassesSelector:
     def select_all_classes(self) -> None:
         self.classes_table.select_all()
 
+    def get_wrong_shape_classes(self, task_type: str) -> List[str]:
+        allowed_shapes = {
+            TaskType.OBJECT_DETECTION: {Rectangle},
+            TaskType.INSTANCE_SEGMENTATION: {Bitmap},
+            TaskType.SEMANTIC_SEGMENTATION: {Bitmap},
+            TaskType.POSE_ESTIMATION: {GraphNodes},
+        }
+
+        if task_type not in allowed_shapes:
+            return []
+
+        selected_classes = self.get_selected_classes()
+        wrong_shape_classes = []
+        for class_name in selected_classes:
+            obj_class = self.classes_table.project_meta.get_obj_class(class_name)
+
+            from supervisely.annotation.obj_class import ObjClass
+
+            obj_class: ObjClass
+            if obj_class is None:
+                continue
+            if obj_class.geometry_type not in allowed_shapes[task_type]:
+                wrong_shape_classes.append(class_name)
+
+        return wrong_shape_classes
+
+    def classify_incompatible_classes(
+        self, task_type: str
+    ) -> Tuple[List[str], List[str]]:
+        """
+        Rules:
+        1) Detection – any shape can be converted, Rectangle is compatible.
+        2) Instance/Semantic segmentation – only Bitmap and Polygon (need conversion to Bitmap) are allowed; other shapes are not convertible.
+        3) Pose estimation – only GraphNodes are allowed; other shapes are not convertible.
+
+        Returns:
+        - convertible: List[str] – list of class names that can be converted to the task type
+        - non_convertible: List[str] – list of class names that cannot be converted to the task type
+        """
+        selected_classes = self.get_selected_classes()
+        convertible: List[str] = []
+        non_convertible: List[str] = []
+        for class_name in selected_classes:
+            obj_class = self.classes_table.project_meta.get_obj_class(class_name)
+            if obj_class is None:
+                continue
+
+            geo_cls = obj_class.geometry_type
+
+            if task_type == TaskType.OBJECT_DETECTION:
+                if geo_cls is Rectangle:
+                    # already compatible
+                    continue
+                # Any other shape is converted to Rectangle (BBox)
+                convertible.append(class_name)
+
+            elif task_type in (
+                TaskType.INSTANCE_SEGMENTATION,
+                TaskType.SEMANTIC_SEGMENTATION,
+            ):
+                if geo_cls is Bitmap:
+                    # already compatible (bitmap mask)
+                    continue
+                if geo_cls is Polygon:
+                    # convertible to bitmap mask
+                    convertible.append(class_name)
+                    continue
+                # Other shapes cannot be converted
+                non_convertible.append(class_name)
+
+            elif task_type == TaskType.POSE_ESTIMATION:
+                if geo_cls is GraphNodes:
+                    # already compatible
+                    continue
+                # Other shapes cannot be converted
+                non_convertible.append(class_name)
+
+            else:
+                # Unknown task type – treat all shapes as compatible
+                continue
+
+        return convertible, non_convertible
+
     def validate_step(self) -> bool:
+        # @TODO: Handle AnyShape classes
         self.validator_text.hide()
+        task_type = (
+            self.model_selector.get_selected_task_type()
+            if self.model_selector
+            else None
+        )
 
-        project_classes = self.classes_table.project_meta.obj_classes
-        if len(project_classes) == 0:
+        if len(self.classes_table.project_meta.obj_classes) == 0:
             self.validator_text.set(text="Project has no classes", status="error")
             self.validator_text.show()
             return False
 
         selected_classes = self.classes_table.get_selected_classes()
-        table_data = self.classes_table._table_data
+        n_classes = len(selected_classes)
+
+        if n_classes == 0:
+            self.validator_text.set(
+                text="Please select at least one class", status="error"
+            )
+            self.validator_text.show()
+            return False
+
+        class_word = "class" if n_classes == 1 else "classes"
+        message_parts = [f"Selected {n_classes} {class_word}"]
+        status = "success"
+        is_valid = True
+
         empty_classes = [
             row[0]["data"]
-            for row in table_data
-            if row[0]["data"] in selected_classes and row[2]["data"] == 0 and row[3]["data"] == 0
+            for row in self.classes_table._table_data
+            if row[0]["data"] in selected_classes
+            and row[2]["data"] == 0
+            and row[3]["data"] == 0
         ]
+        if empty_classes:
+            empty_word = "class" if len(empty_classes) == 1 else "classes"
+            message_parts.append(
+                f"{empty_word.capitalize()} with no annotations: {', '.join(empty_classes)}"
+            )
+            status = "warning"
 
-        n_classes = len(selected_classes)
-        if n_classes == 0:
-            message = "Please select at least one class"
-            status = "error"
-        else:
-            class_text = "class" if n_classes == 1 else "classes"
-            message = f"Selected {n_classes} {class_text}"
-            status = "success"
-            if empty_classes:
-                intersections = set(selected_classes).intersection(empty_classes)
-                if intersections:
-                    class_text = "class" if len(intersections) == 1 else "classes"
-                    message += (
-                        f". Selected {class_text} have no annotations: {', '.join(intersections)}"
+        convertible_classes, non_convertible_classes = (
+            self.classify_incompatible_classes(task_type)
+        )
+
+        if (
+            n_classes > 0
+            and not convertible_classes
+            and len(non_convertible_classes) == n_classes
+        ):
+            available_task_types = []
+            if self.model_selector is not None and hasattr(
+                self.model_selector, "pretrained_models_table"
+            ):
+                available_task_types = (
+                    self.model_selector.pretrained_models_table.get_available_task_types()
+                )
+
+            other_suitable_exists = False
+            for other_task_type in available_task_types:
+                if other_task_type == task_type:
+                    continue
+                other_conv, other_non_conv = self.classify_incompatible_classes(
+                    other_task_type
+                )
+                if other_conv or len(other_non_conv) < n_classes:
+                    other_suitable_exists = True
+                    break
+
+            if other_suitable_exists:
+                self.validator_text.set(
+                    text=(
+                        f"No suitable classes for task type {task_type}. "
+                        "Training is not possible. Please choose another task type."
+                    ),
+                    status="error",
+                )
+            else:
+                self.validator_text.set(
+                    text=(
+                        "Project contains no suitable classes for training. "
+                        "Training is not possible. Please select another project."
+                    ),
+                    status="error",
+                )
+            self.validator_text.show()
+            return False
+
+        incompatible_exist = bool(convertible_classes or non_convertible_classes)
+        if incompatible_exist:
+            task_specific_texts = {
+                TaskType.OBJECT_DETECTION: "Only rectangle shapes are supported for object detection task, all other shape types can be converted to rectangle",
+                TaskType.INSTANCE_SEGMENTATION: "Only bitmap shapes are supported for instance segmentation task, polygon shapes can be converted to bitmap",
+                TaskType.SEMANTIC_SEGMENTATION: "Only bitmap shapes are supported for semantic segmentation task, polygon shapes can be converted to bitmap",
+                TaskType.POSE_ESTIMATION: "Only keypoint (graph) shape is supported for pose estimation task",
+            }
+
+            if non_convertible_classes:
+                specific_text = task_specific_texts.get(
+                    task_type,
+                    "Some selected classes have shapes that are incompatible with the chosen model task type.",
+                )
+                message_parts = [f"Model task type is {task_type}. {specific_text}"]
+                message_parts.append("Remove incompatible classes to continue.")
+                status = "error"
+                is_valid = False
+            else:
+                if self.is_convert_class_shapes_enabled():
+                    message_parts.append(
+                        f"Conversion enabled. Classes will be converted for task {task_type}."
+                    )
+                    status = "info" if status == "success" else status
+                    is_valid = True
+                else:
+                    specific_text = task_specific_texts.get(
+                        task_type,
+                        "Some selected classes have shapes that are incompatible with the chosen model task type.",
                     )
-                    status = "warning"
+                    message_parts = [
+                        f"Model task type is {task_type}. {specific_text}",
+                        "Enable conversion checkbox or select compatible classes to continue.",
+                    ]
+                    status = "error"
+                    is_valid = False
+        else:
+            if self.is_convert_class_shapes_enabled():
+                message_parts.append(
+                    "Conversion enabled, but no shape conversion required."
+                )
 
-        self.validator_text.set(text=message, status=status)
+        self.validator_text.set(text=". ".join(message_parts), status=status)
         self.validator_text.show()
-        return n_classes > 0
+        return is_valid
+
+    def is_convert_class_shapes_enabled(self) -> bool:
+        return self.convert_class_shapes_checkbox.is_checked()