supervisely 6.73.410__py3-none-any.whl → 6.73.470__py3-none-any.whl

supervisely/nn/inference/predict_app/gui/settings_selector.py

@@ -0,0 +1,881 @@
+import os
+import random
+import shutil
+import subprocess
+import threading
+from contextlib import contextmanager, nullcontext
+from pathlib import Path
+from typing import Any, Callable, Dict, List
+
+import cv2
+import yaml
+
+from supervisely._utils import logger
+from supervisely.annotation.annotation import Annotation
+from supervisely.annotation.label import Label
+from supervisely.api.api import Api
+from supervisely.api.video.video_api import VideoInfo
+from supervisely.app.widgets import (
+    Button,
+    Card,
+    Container,
+    Editor,
+    Field,
+    GridGallery,
+    Input,
+    InputNumber,
+    OneOf,
+    Progress,
+    Select,
+    Text,
+    VideoPlayer,
+)
+from supervisely.app.widgets.checkbox.checkbox import Checkbox
+from supervisely.app.widgets.empty.empty import Empty
+from supervisely.app.widgets.widget import Widget
+from supervisely.nn.inference.inference import (
+    _filter_duplicated_predictions_from_ann,
+    update_meta_and_ann,
+    update_meta_and_ann_for_video_annotation,
+)
+from supervisely.nn.inference.predict_app.gui.input_selector import InputSelector
+from supervisely.nn.inference.predict_app.gui.model_selector import ModelSelector
+from supervisely.nn.inference.predict_app.gui.utils import (
+    video_annotation_from_predictions,
+)
+from supervisely.nn.model.model_api import ModelAPI, Prediction
+from supervisely.nn.tracker import TrackingVisualizer
+from supervisely.project import ProjectMeta
+from supervisely.project.project_meta import ProjectType
+from supervisely.video.video import VideoFrameReader
+from supervisely.video_annotation.video_annotation import KeyIdMap, VideoAnnotation
+
+
+class InferenceMode:
+    FULL_IMAGE = "Full Image"
+    SLIDING_WINDOW = "Sliding Window"
+
+
+class AddPredictionsMode:
+    APPEND = "Merge with existing labels"
+    REPLACE = "Replace existing labels"
+    IOU_MERGE = "Merge by IoU threshold"
+    REPLACE_EXISTING_LABELS_AND_SAVE_IMAGE_TAGS = "Replace existing labels and save image tags"
+
+
+class Preview:
+    lock_message = "Select previous step to unlock"
+
+    def __init__(
+        self,
+        api: Api,
+        preview_dir: str,
+        get_model_api_fn: Callable[[], ModelAPI],
+        get_input_settings_fn: Callable[[], Dict[str, Any]],
+        get_settings_fn: Callable[[], Dict[str, Any]],
+    ):
+        self.api = api
+        self.preview_dir = preview_dir
+        self.get_model_api_fn = get_model_api_fn
+        self.get_input_settings_fn = get_input_settings_fn
+        self.get_settings_fn = get_settings_fn
+        os.makedirs(self.preview_dir, exist_ok=True)
+        os.makedirs(Path(self.preview_dir, "annotated"), exist_ok=True)
+        self.image_preview_path = None
+        self.image_peview_url = None
+        self.video_preview_path = None
+        self.video_preview_annotated_path = None
+        self.video_peview_url = None
+
+        self.progress_widget = Progress(show_percents=True, hide_on_finish=True)
+        self.download_error = Text("", status="warning")
+        self.download_error.hide()
+        self.progress_container = Container(widgets=[self.download_error, self.progress_widget])
+        self.loading_container = Container(widgets=[self.download_error, Text("Loading...")])
+
+        self.image_gallery = GridGallery(
+            2,
+            sync_views=True,
+            enable_zoom=True,
+            resize_on_zoom=True,
+            empty_message="",
+        )
+        self.image_preview_container = Container(widgets=[self.image_gallery])
+
+        self.video_player = VideoPlayer()
+        self.video_preview_container = Container(widgets=[self.video_player])
+
+        self.locked_text = Text("Select input and model to unlock", status="info")
+        self.empty_text = Text("Click preview to visualize predictions")
+        self.error_text = Text("Failed to generate preview", status="error")
+
+        self.select = Select(
+            items=[
+                Select.Item("locked", content=self.locked_text),
+                Select.Item("empty", content=self.empty_text),
+                Select.Item(ProjectType.IMAGES.value, content=self.image_preview_container),
+                Select.Item(ProjectType.VIDEOS.value, content=self.video_preview_container),
+                Select.Item("error", content=self.error_text),
+                Select.Item("loading", content=self.loading_container),
+                Select.Item("progress", content=self.progress_container),
+            ]
+        )
+        self.select.set_value("empty")
+        self.oneof = OneOf(self.select)
+
+        self.run_button = Button("Preview", icon="zmdi zmdi-slideshow")
+        self.run_button.disable()
+        self.card = Card(
+            title="Preview",
+            description="Preview model predictions on a random image or video from the selected input source.",
+            content=self.oneof,
+            content_top_right=self.run_button,
+            lock_message=self.lock_message,
+        )
+
+        @self.run_button.click
+        def _run_preview():
+            self.run_preview()
+
+    def lock(self):
+        self.run_button.disable()
+        self.card.lock(self.lock_message)
+
+    def unlock(self):
+        self.run_button.enable()
+        self.card.unlock()
+
+    @contextmanager
+    def progress(self, message: str, total: int, **kwargs):
+        current_item = self.select.get_value()
+        try:
+            with self.progress_widget(message=message, total=total, **kwargs) as pbar:
+                self.select_item("progress")
+                yield pbar
+        finally:
+            self.select_item(current_item)
+
+    def select_item(self, item: str):
+        self.select.set_value(item)
+
+    def _download_video_by_frames(
+        self, video_info: VideoInfo, save_path: str, frames_number=150, progress_cb=None
+    ):
+        if Path(save_path).exists():
+            Path(save_path).unlink()
+        tmp_dir = Path(self.preview_dir, "tmp_frames")
+        if tmp_dir.exists():
+            shutil.rmtree(tmp_dir)
+        os.makedirs(tmp_dir, exist_ok=True)
+        self.api.video.download_frames(
+            video_info.id,
+            frames=list(range(frames_number)),
+            paths=[str(tmp_dir / f"frame_{i}.jpg") for i in range(frames_number)],
+            progress_cb=progress_cb,
+        )
+        fps = int(video_info.frames_count / video_info.duration)
+        fourcc = cv2.VideoWriter.fourcc(*"mp4v")  # or 'avc1', 'XVID', 'H264'
+        out = cv2.VideoWriter(
+            save_path, fourcc, fps, (video_info.frame_width, video_info.frame_height)
+        )
+        for i in range(frames_number):
+            frame_path = tmp_dir / f"frame_{i}.jpg"
+            if not frame_path.exists():
+                continue
+            img = cv2.imread(str(frame_path))
+            out.write(img)
+        out.release()
+        shutil.rmtree(tmp_dir)
+
+    def _download_full_video(
+        self, video_id: int, save_path: str, duration: int = 5, progress_cb=None
+    ):
+        if Path(save_path).exists():
+            Path(save_path).unlink()
+        temp = Path(self.preview_dir) / f"temp_{video_id}.mp4"
+        if temp.exists():
+            temp.unlink()
+        self.api.video.download_path(video_id, temp, progress_cb=progress_cb)
+        minutes = duration // 60
+        hours = minutes // 60
+        minutes = minutes % 60
+        seconds = duration % 60
+        duration_str = f"{hours:02}:{minutes:02}:{seconds:02}"
+        try:
+            process = subprocess.Popen(
+                [
+                    "ffmpeg",
+                    "-y",
+                    "-i",
+                    str(temp),
+                    "-c",
+                    "copy",
+                    "-t",
+                    duration_str,
+                    save_path,
+                ],
+                stderr=subprocess.PIPE,
+            )
+            process.wait()
+            logger.debug("FFmpeg exited with code: " + str(process.returncode))
+            logger.debug(f"FFmpeg stderr: {process.stderr.read().decode()}")
+            if len(VideoFrameReader(save_path).read_frames()) == 0:
+                raise RuntimeError("No frames read from the video")
+            temp.unlink()
+        except Exception as e:
+            if Path(save_path).exists():
+                Path(save_path).unlink()
+            shutil.copy(temp, save_path)
+            temp.unlink()
+            logger.warning(f"FFmpeg trimming failed: {str(e)}", exc_info=True)
+
+    def _download_video_preview(self, video_info: VideoInfo, with_progress=True):
+        video_id = video_info.id
+        duration = 5
+        video_path = Path(self.preview_dir, video_info.name)
+        self.video_preview_path = video_path
+        self.video_preview_annotated_path = Path(self.preview_dir, "annotated") / Path(
+            self.video_preview_path
+        ).relative_to(self.preview_dir)
+        success = False
+        try:
+            try:
+                size = int(video_info.file_meta["size"])
+                size = int(size / video_info.duration * duration)
+            except:
+                size = None
+            with (
+                self.progress("Downloading video part:", total=size, unit="B", unit_scale=True)
+                if with_progress and size
+                else nullcontext()
+            ) as pbar:
+                success = self._partial_download(
+                    video_id, duration, str(self.video_preview_path), progress_cb=pbar.update
+                )
+        except Exception as e:
+            logger.warning(f"Partial download failed: {str(e)}", exc_info=True)
+            success = False
+        if success:
+            return
+
+        video_length_threshold = 120  # seconds
+        if video_info.duration > video_length_threshold:
+            self.download_error.text = (
+                f"Partial download failed. Will Download separate video frames"
+            )
+            self.download_error.show()
+
+            fps = int(video_info.frames_count / video_info.duration)
+            frames_number = min(video_info.frames_count, int(fps * duration))
+            with (
+                self.progress(
+                    "Downloading video frames:", total=frames_number, unit="it", unit_scale=False
+                )
+                if with_progress
+                else nullcontext()
+            ) as pbar:
+                self._download_video_by_frames(
+                    video_info,
+                    str(self.video_preview_path),
+                    frames_number=frames_number,
+                    progress_cb=pbar.update,
+                )
+        else:
+            self.download_error.text = f"Partial download failed. Will Download full video"
+            self.download_error.show()
+            size = int(video_info.file_meta["size"])
+            with (
+                self.progress("Downloading video:", total=size, unit="B", unit_scale=True)
+                if with_progress
+                else nullcontext()
+            ) as pbar:
+                self._download_full_video(
+                    video_info.id,
+                    str(self.video_preview_path),
+                    duration=duration,
+                    progress_cb=pbar.update,
+                )
+
+    def _partial_download(self, video_id: int, duration: int, save_path: str, progress_cb=None):
+        if Path(save_path).exists():
+            Path(save_path).unlink()
+        duration_minutes = duration // 60
+        duration_hours = duration_minutes // 60
+        duration_minutes = duration_minutes % 60
+        duration_seconds = duration % 60
+        duration_str = f"{duration_hours:02}:{duration_minutes:02}:{duration_seconds:02}"
+        response = self.api.video._download(video_id, is_stream=True)
+        process = subprocess.Popen(
+            [
+                "ffmpeg",
+                "-y",
+                "-t",
+                duration_str,
+                "-probesize",
+                "50M",
+                "-analyzeduration",
+                "50M",
+                "-i",
+                "pipe:0",
+                "-movflags",
+                "frag_keyframe+empty_moov+default_base_moof",
+                "-c",
+                "copy",
+                save_path,
+            ],
+            stdin=subprocess.PIPE,
+            stderr=subprocess.PIPE,
+        )
+
+        bytes_written = 0
+        try:
+            for chunk in response.iter_content(chunk_size=8192):
+                process.stdin.write(chunk)
+                bytes_written += len(chunk)
+                if progress_cb:
+                    progress_cb(len(chunk))
+        except (BrokenPipeError, IOError):
+            logger.debug("FFmpeg process closed the pipe, stopping download.", exc_info=True)
+            pass
+        finally:
+            process.stdin.close()
+            process.wait()
+            response.close()
+            logger.debug("FFmpeg exited with code: " + str(process.returncode))
+            logger.debug(f"FFmpeg stderr: {process.stderr.read().decode()}")
+            logger.debug(f"Total bytes written: {bytes_written}")
+        try:
+            with VideoFrameReader(save_path) as reader:
+                if len(reader.read_frames()) == 0:
+                    return False
+            return True
+        except Exception as e:
+            return False
+
+    def _download_preview_item(self, with_progress: bool = False):
+        input_settings = self.get_input_settings_fn()
+        video_ids = input_settings.get("video_ids", None)
+        if video_ids is None:
+            project_id = input_settings.get("project_id", None)
+            dataset_ids = input_settings.get("dataset_ids", None)
+            if dataset_ids:
+                images = []
+                candidate_ids = list(dataset_ids)
+                random.shuffle(candidate_ids)
+                dataset_id = None
+                for ds_id in candidate_ids:
+                    images = self.api.image.get_list(ds_id)
+                    if images:
+                        dataset_id = ds_id
+                        break
+                if not images:
+                    raise RuntimeError("No images found in the selected datasets")
+            else:
+                datasets = self.api.dataset.get_list(project_id)
+                total_items = sum(ds.items_count for ds in datasets)
+                if total_items == 0:
+                    raise RuntimeError("No images found in the selected datasets")
+                images = []
+                while not images:
+                    dataset_id = random.choice(datasets).id
+                    images = self.api.image.get_list(dataset_id)
+            image_id = random.choice(images).id
+            image_info = self.api.image.get_info_by_id(image_id)
+            self.image_preview_path = Path(self.preview_dir, image_info.name)
+            self.api.image.download_path(image_id, self.image_preview_path)
+            self._current_item_id = image_id
+            ann_info = self.api.annotation.download(image_id)
+            self._project_meta = ProjectMeta.from_json(
+                self.api.project.get_meta(image_info.project_id)
+            )
+            self._image_annotation = Annotation.from_json(ann_info.annotation, self._project_meta)
+            self.image_peview_url = f"./static/preview/{image_info.name}"
+        elif len(video_ids) == 0:
+            self._current_item_id = None
+            self.video_preview_path = None
+            self.video_peview_url = None
+            self.video_preview_annotated_path = None
+        else:
+            video_id = random.choice(video_ids)
+            video_id = video_ids[0]
+            video_info = self.api.video.get_info_by_id(video_id)
+            self._download_video_preview(video_info, with_progress)
+            self._current_item_id = video_id
+            self.video_peview_url = f"./static/preview/annotated/{video_info.name}"
+            self._project_meta = ProjectMeta.from_json(
+                self.api.project.get_meta(video_info.project_id)
+            )
+            self._video_annotation = VideoAnnotation.from_json(
+                self.api.video.annotation.download(video_id), self._project_meta, KeyIdMap()
+            )
+
+    def set_image_preview(self):
+
+        def _maybe_merge_annotations(
+            source: Annotation,
+            pred: Annotation,
+            predictions_mode: str,
+            model_prediction_suffix: str,
+            iou_threshold: float = None,
+        ):
+            project_meta, pred, _ = update_meta_and_ann(
+                self._project_meta, pred, model_prediction_suffix
+            )
+            if predictions_mode == AddPredictionsMode.REPLACE:
+                return pred
+            elif predictions_mode == AddPredictionsMode.IOU_MERGE:
+                iou_threshold = iou_threshold if iou_threshold is not None else 0.9
+                pred = _filter_duplicated_predictions_from_ann(source, pred, iou_threshold)
+                return source.merge(pred)
+            elif predictions_mode in [
+                AddPredictionsMode.APPEND,
+                AddPredictionsMode.REPLACE_EXISTING_LABELS_AND_SAVE_IMAGE_TAGS,
+            ]:
+                return source.merge(pred)
+            else:
+                raise RuntimeError(f"Unknown predictions mode: {predictions_mode}")
+
+        self.image_gallery.clean_up()
+        if not self._current_item_id:
+            self._download_preview_item(with_progress=True)
+        image_id = self._current_item_id
+        model_api = self.get_model_api_fn()
+        settings = self.get_settings_fn()
+        inference_settings = settings.get("inference_settings", {})
+        with self.progress("Running Model:", total=1) as pbar:
+            prediction = model_api.predict(
+                image_id=image_id, inference_settings=inference_settings, tqdm=pbar
+            )[0]
+            prediction_annotation = _maybe_merge_annotations(
+                source=self._image_annotation,
+                pred=prediction.annotation,
+                predictions_mode=settings.get("predictions_mode", AddPredictionsMode.APPEND),
+                model_prediction_suffix=settings.get("model_prediction_suffix", ""),
+                iou_threshold=inference_settings.get("existing_objects_iou_thresh"),
+            )
+        self.image_gallery.append(
+            self.image_peview_url, title="Source", annotation=self._image_annotation
+        )
+        self.image_gallery.append(
+            self.image_peview_url, title="Prediction", annotation=prediction_annotation
+        )
+        self.select_item(ProjectType.IMAGES.value)
+
+    def set_video_preview(
+        self,
+    ):
+        self.video_player.set_video(None)
+        input_settings = self.get_input_settings_fn()
+        video_ids = input_settings.get("video_ids", None)
+        if not video_ids:
+            raise RuntimeError("No videos selected")
+        if not self._current_item_id:
+            self._download_preview_item(with_progress=True)
+        video_id = self._current_item_id
+
+        frame_start = 0
+        seconds = 5
+        video_info = self.api.video.get_info_by_id(video_id)
+        fps = int(video_info.frames_count / video_info.duration)
+        frames_number = min(video_info.frames_count, int(fps * seconds))
+        model_api = self.get_model_api_fn()
+        project_meta = ProjectMeta.from_json(self.api.project.get_meta(video_info.project_id))
+
+        settings = self.get_settings_fn()
+        inference_settings = settings.get("inference_settings", {})
+        tracking = settings.get("tracking", False)
+        with self.progress("Running model:", total=frames_number) as pbar:
+            with model_api.predict_detached(
+                video_id=video_id,
+                inference_settings=inference_settings,
+                tracking=tracking,
+                start_frame=frame_start,
+                num_frames=frames_number,
+                tqdm=pbar,
+            ) as session:
+                predictions: List[Prediction] = list(session)
+
+        if os.path.exists(self.video_preview_annotated_path):
+            os.remove(self.video_preview_annotated_path)
+        if tracking:
+            pred_video_annotation = session.final_result.get("video_ann", {})
+            if pred_video_annotation is None:
+                raise RuntimeError("Model did not return video annotation")
+            pred_video_annotation = VideoAnnotation.from_json(
+                pred_video_annotation, project_meta=project_meta
+            )
+            _, pred_video_annotation, _ = update_meta_and_ann_for_video_annotation(
+                self._project_meta,
+                pred_video_annotation,
+                settings.get("model_prediction_suffix", ""),
+            )
+            visualizer = TrackingVisualizer(
+                output_fps=fps,
+                box_thickness=video_info.frame_height // 110,
+                text_scale=video_info.frame_height / 900,
+                trajectory_thickness=video_info.frame_height // 110,
+            )
+        else:
+            pred_video_annotation = video_annotation_from_predictions(
+                predictions,
+                model_api.get_model_meta(),
+                frame_size=(video_info.frame_height, video_info.frame_width),
+            )
+            visualizer = TrackingVisualizer(
+                output_fps=fps,
+                box_thickness=video_info.frame_height // 110,
+                text_scale=video_info.frame_height / 900,
+                show_trajectories=False,
+            )
+        _, pred_video_annotation, _ = update_meta_and_ann_for_video_annotation(
+            self._project_meta,
+            pred_video_annotation,
+            settings.get("model_prediction_suffix", ""),
+        )
+        visualizer.visualize_video_annotation(
+            pred_video_annotation,
+            source=self.video_preview_path,
+            output_path=self.video_preview_annotated_path,
+        )
+        self.video_player.set_video(self.video_peview_url)
+        self.select_item(ProjectType.VIDEOS.value)
+
+    def set_error(self, text: str):
+        self.error_text.text = text
+        self.select_item("error")
+
+    def run_preview(self):
+        self.download_error.hide()
+        self.select_item("loading")
+        try:
+            input_settings = self.get_input_settings_fn()
+            video_ids = input_settings.get("video_ids", None)
+            if video_ids is None:
+                self.set_image_preview()
+            elif len(video_ids) == 0:
+                self.set_error("No videos selected")
+            else:
+                self.set_video_preview()
+        except Exception as e:
+            logger.error(f"Failed to generate preview: {str(e)}", exc_info=True)
+            self.set_error("Failed to generate preview: " + str(e))
+
+    def _preload_item(self):
+        threading.Thread(
+            target=self._download_preview_item, kwargs={"with_progress": False}, daemon=True
+        ).start()
+
+    def update_item_type(self, item_type: str):
+        self.select_item("empty")
+        self._current_item_id = None
+        self.download_error.hide()
+        # self._preload_item() # need to handle race condition with run_preview and multiple clicks
+
+
+class SettingsSelector:
+    title = "Inference (settings + preview)"
+    description = "Select additional settings for model inference"
+    lock_message = "Select previous step to unlock"
+
+    def __init__(
+        self,
+        api: Api,
+        static_dir: str,
+        input_selector: InputSelector,
+        model_selector: ModelSelector,
+    ):
+        # Init Step
+        self.api = api
+        self.static_dir = static_dir
+        self.input_selector = input_selector
+        self.model_selector = model_selector
+        self.display_widgets: List[Any] = []
+        # -------------------------------- #
+
+        # Init Base Widgets
+        self.validator_text = None
+        self.button = None
+        self.run_button = None
+        self.container = None
+        self.cards = None
+        # -------------------------------- #
+
+        # Init Step Widgets
+        self.inference_mode_selector = None
+        self.inference_mode_field = None
+        self.model_prediction_suffix_input = None
+        self.model_prediction_suffix_field = None
+        # self.model_prediction_suffix_checkbox = None
+        self.predictions_mode_selector = None
+        self.predictions_mode_field = None
+        self.inference_settings_editor = None
+        # -------------------------------- #
+
+        self.settings_widgets = []
+        self.image_settings_widgets = []
+        self.video_settings_widgets = []
+
+        # Prediction Mode
+        self.prediction_modes = [
+            AddPredictionsMode.APPEND,
+            AddPredictionsMode.REPLACE,
+            AddPredictionsMode.IOU_MERGE,
+            # AddPredictionsMode.REPLACE_EXISTING_LABELS_AND_SAVE_IMAGE_TAGS, # @TODO: Implement later
+        ]
+        self.iou_merge_input = InputNumber(value=0.9, min=0.0, max=1.0, step=0.05, controls=False)
+        self.iou_merge_input_field = Field(
+            content=self.iou_merge_input,
+            title="IoU Threshold",
+            description="IoU threshold for merging predictions with existing labels. Predictions with IoU above this threshold will be considered duplicates and removed.",
+        )
+        self.prediction_modes_contents = [Empty(), Empty(), self.iou_merge_input_field]
+        self.predictions_mode_selector = Select(
+            items=[
+                Select.Item(mode, content=content)
+                for mode, content in zip(self.prediction_modes, self.prediction_modes_contents)
+            ]
+        )
+        self.predictions_mode_selector.set_value(self.prediction_modes[0])
+        self.predicitons_mode_one_of = OneOf(self.predictions_mode_selector)
+        self.predictions_mode_field = Field(
+            content=Container(
+                widgets=[self.predictions_mode_selector, self.predicitons_mode_one_of]
+            ),
+            title="Add predictions mode",
+            description="Select how to add predictions to the project: by merging with existing labels or by replacing them.",
+        )
+        # Add widgets to display ------------ #
+        self.image_settings_widgets.extend([self.predictions_mode_field])
+        # ----------------------------------- #
+
+        # Tracking
+        self.tracking_checkbox = Checkbox(content="Enable tracking", checked=True)
+        self.tracking_checkbox_field = Field(
+            content=self.tracking_checkbox,
+            title="Tracking",
+            description="Enable tracking for video predictions. The tracking algorithm is BoT-SORT version improved by Supervisely team.",
+        )
+        # Add widgets to display ------------ #
+        self.video_settings_widgets.extend([self.tracking_checkbox_field])
+        self.image_settings_container = Container(widgets=self.image_settings_widgets, gap=15)
+        self.video_settings_container = Container(widgets=self.video_settings_widgets, gap=15)
+        self.image_or_video_container = Container(
+            widgets=[self.image_settings_container, self.video_settings_container], gap=0
+        )
+        self.video_settings_container.hide()
+        self.settings_widgets.extend([self.image_or_video_container])
+        # ----------------------------------- #
+
+        # Class / Tag Suffix
+        self.model_prediction_suffix_input = Input(
+            value="_model", minlength=1, placeholder="Enter suffix e.g: _model"
+        )
+        self.model_meta_has_conflicting_names_text = Text(
+            text="Project and Model metas have conflicting names. This suffix will be added to conflicting class and tag names of model predictions",
+            status="info",
+        )
+        self.model_prediction_suffix_field = Field(
+            content=self.model_prediction_suffix_input,
+            title="Class and tag suffix",
+            description=(
+                "Suffix that will be added to conflicting class and tag names. "
+                "E.g. your project has a class 'person' with shape 'bitmap' and model has class 'person' with shape 'rectangle', "
+                "then suffix will be added to the model predictions to avoid conflicts. E.g. 'person_model'."
+            ),
+        )
+        self.model_prediction_suffix_container = Container(
+            widgets=[
+                self.model_meta_has_conflicting_names_text,
+                self.model_prediction_suffix_field,
+            ],
+            gap=5,
+        )
+        self.model_prediction_suffix_container.hide()
+        # Add widgets to display ------------ #
+        self.settings_widgets.extend([self.model_prediction_suffix_container])
+        # ----------------------------------- #
+
+        # Inference Settings
+        self.inference_settings_editor = Editor("", language_mode="yaml", height_px=300)
+        self.inference_settings_field = Field(
+            content=self.inference_settings_editor,
+            title="Inference and Tracking Settings",
+        )
+        # Add widgets to display ------------ #
+        self.settings_widgets.extend([self.inference_settings_field])
+        # ----------------------------------- #
+
+        # Preview
+        self.preview_dir = os.path.join(self.static_dir, "preview")
+        self.preview = Preview(
+            api=self.api,
+            preview_dir=self.preview_dir,
+            get_model_api_fn=lambda: self.model_selector.model.model_api,
+            get_input_settings_fn=self.input_selector.get_settings,
+            get_settings_fn=self.get_settings,
+        )
+
+        self.settings_container = Container(widgets=self.settings_widgets, gap=15)
+        self.display_widgets.extend([self.settings_container])
+        # Base Widgets
+        self.validator_text = Text("")
+        self.validator_text.hide()
+        self.button = Button("Select")
+        # Add widgets to display ------------ #
+        self.display_widgets.extend([self.validator_text, self.button])
+        # ----------------------------------- #
+
+        # Card Layout
+        self.container = Container(self.display_widgets)
+        self.settings_card = Card(
+            title=self.title,
+            description=self.description,
+            content=self.container,
+            lock_message=self.lock_message,
+        )
+        self.cards = [self.settings_card, self.preview.card]
+        self.cards_container = Container(
+            widgets=self.cards,
+            gap=15,
+            direction="horizontal",
+            fractions=[3, 7],
+        )
+        # ----------------------------------- #
+
+    def lock(self):
+        self.settings_card.lock(self.lock_message)
+        self.preview.lock()
+
+    def unlock(self):
+        self.settings_card.unlock()
+        self.preview.unlock()
+
+    def disable(self):
+        for widget in self.widgets_to_disable:
+            if isinstance(widget, Editor):
+                widget.readonly = True
+            else:
+                widget.disable()
+
+    def enable(self):
+        for widget in self.widgets_to_disable:
+            if isinstance(widget, Editor):
+                widget.readonly = False
+            else:
+                widget.enable()
+
+    @property
+    def widgets_to_disable(self) -> List[Widget]:
+        return [
+            self.tracking_checkbox,
+            self.predictions_mode_selector,
+            self.model_prediction_suffix_input,
+            self.inference_settings_editor,
+        ]
+
+    def set_inference_settings(self, settings: Dict[str, Any]):
+        settings = "# Inference settings\n" + settings
+        if isinstance(settings, str):
+            self.inference_settings_editor.set_text(settings)
+        else:
+            self.inference_settings_editor.set_text(yaml.safe_dump(settings))
+
+    def set_tracking_settings(self, settings: Dict[str, Any]):
+        if self.input_selector.radio.get_value() != ProjectType.VIDEOS.value:
+            return
+
+        current_settings = self.inference_settings_editor.get_text()
+        if isinstance(settings, str):
+            all_settings = current_settings + "\n\n# Tracking settings\n" + settings
+            self.inference_settings_editor.set_text(all_settings)
+        else:
+            all_settings = current_settings + "\n\n# Tracking settings\n" + yaml.safe_dump(settings)
+            self.inference_settings_editor.set_text(all_settings)
+
+    def set_default_tracking_settings(self):
+        nn_dir = Path(__file__).parents[3]
+        config_path = nn_dir / "tracker" / "botsort" / "botsort_config.yaml"
+
+        with open(config_path, "r", encoding="utf-8") as file:
+            botsort_config = yaml.safe_load(file)
+        self.set_tracking_settings(botsort_config)
+
+    def get_inference_settings(self) -> Dict:
+        text = self.inference_settings_editor.get_text()
+        inference_settings_text = text.split("# Tracking settings")[0]
+        settings = yaml.safe_load(inference_settings_text)
+        settings = settings if settings is not None else {}
+        if (
+            self.input_selector.radio.get_value() == ProjectType.IMAGES.value
+            and self.predictions_mode_selector.get_value() == AddPredictionsMode.IOU_MERGE
+        ):
+            settings["existing_objects_iou_thresh"] = self.iou_merge_input.get_value()
+        return settings
+
+    def get_tracking_settings(self) -> Dict:
+        if self.input_selector.radio.get_value() != ProjectType.VIDEOS.value:
+            return {}
+
+        text = self.inference_settings_editor.get_text()
+        text_parts = text.split("# Tracking settings")
+        if len(text_parts) > 1:
+            tracking_settings_text = text_parts[1]
+            settings = yaml.safe_load(tracking_settings_text)
+            if settings:
+                return settings
+        return {}
+
+    def get_settings(self) -> Dict[str, Any]:
+        settings = {
+            # "inference_mode": self.inference_mode_selector.get_value(),
+            "inference_mode": InferenceMode.FULL_IMAGE,
+            "model_prediction_suffix": self.model_prediction_suffix_input.get_value(),
+            "predictions_mode": self.predictions_mode_selector.get_value(),
+            "inference_settings": self.get_inference_settings(),
+        }
+        if self.input_selector.radio.get_value() == ProjectType.VIDEOS.value:
+            settings["tracking_settings"] = self.get_tracking_settings()
+            if self.input_selector.get_settings().get("video_ids", None) is not None:
+                settings["tracking"] = self.tracking_checkbox.is_checked()
+        return settings
+
+    def load_from_json(self, data):
+        # inference_mode = data.get("inference_mode", None)
+        # if inference_mode:
+        #     self.inference_mode_selector.set_value(inference_mode)
+
+        model_prediction_suffix = data.get("model_prediction_suffix", None)
+        if model_prediction_suffix is not None:
+            self.model_prediction_suffix_input.set_value(model_prediction_suffix)
+
+        predictions_mode = data.get("predictions_mode", None)
+        if predictions_mode:
+            self.predictions_mode_selector.set_value(predictions_mode)
+
+        inference_settings = data.get("inference_settings", None)
+        if inference_settings is not None:
+            self.set_inference_settings(inference_settings)
+
+        tracking_settings = data.get("tracking_settings", None)
+        if tracking_settings is not None:
+            self.set_tracking_settings(tracking_settings)
+
+        tracking = data.get("tracking", None)
+        if tracking == True:
+            self.tracking_checkbox.check()
+        elif tracking == False:
+            self.tracking_checkbox.uncheck()
+
+    def update_item_type(self, item_type: str):
+        if item_type == ProjectType.IMAGES.value:
+            self.video_settings_container.hide()
+            self.image_settings_container.show()
+        elif item_type == ProjectType.VIDEOS.value:
+            self.image_settings_container.hide()
+            self.video_settings_container.show()
+        else:
+            raise ValueError(f"Unsupported item type: {item_type}")
+        self.preview.update_item_type(item_type)
+
+    def validate_step(self) -> bool:
+        return True