supervisely 6.73.427__py3-none-any.whl → 6.73.429__py3-none-any.whl

supervisely/nn/tracker/visualize.py

@@ -0,0 +1,519 @@
+from typing import Union, Dict, List, Tuple, Iterator, Optional
+import numpy as np
+import cv2
+import ffmpeg
+from pathlib import Path
+from collections import defaultdict
+from dataclasses import dataclass
+import tempfile
+import shutil
+
+import supervisely as sly
+from supervisely import logger
+from supervisely.nn.model.prediction import Prediction
+from supervisely import VideoAnnotation
+from supervisely.nn.tracker.utils import predictions_to_video_annotation
+
+
+class TrackingVisualizer:
+    def __init__(
+        self,
+        show_labels: bool = True,
+        show_classes: bool = True,
+        show_trajectories: bool = True,
+        show_frame_number: bool = False,
+        box_thickness: int = 2,
+        text_scale: float = 0.6,
+        text_thickness: int = 2,
+        trajectory_length: int = 30,
+        codec: str = "mp4",
+        output_fps: float = 30.0,
+        colorize_tracks: bool = True,
+
+    ):
+        """
+        Initialize the visualizer with configuration.
+
+        Args:
+            show_labels: Whether to show track IDs.
+            show_classes: Whether to show class names.
+            show_trajectories: Whether to draw trajectories.
+            show_frame_number: Whether to overlay frame number.
+            box_thickness: Thickness of bounding boxes.
+            text_scale: Scale of label text.
+            text_thickness: Thickness of label text.
+            trajectory_length: How many points to keep in trajectory.
+            codec: Output video codec.
+            output_fps: Output video framerate.
+            colorize_tracks (bool, default=True): if True, ignore colors from project meta and generate new colors for each tracked object; if False, try to use colors from project meta when possible.
+        """
+        # Visualization settings
+        self.show_labels = show_labels
+        self.show_classes = show_classes
+        self.show_trajectories = show_trajectories
+        self.show_frame_number = show_frame_number
+
+        # Style settings
+        self.box_thickness = box_thickness
+        self.text_scale = text_scale
+        self.text_thickness = text_thickness
+        self.trajectory_length = trajectory_length
+        self.colorize_tracks = colorize_tracks
+
+        # Output settings
+        self.codec = codec
+        self.output_fps = output_fps
+
+        # Internal state
+        self.annotation = None
+        self.tracks_by_frame = {}
+        self.track_centers = defaultdict(list)
+        self.track_colors = {}
+        self.color_palette = self._generate_color_palette()
+        self._temp_dir = None
+        
+    def _generate_color_palette(self, num_colors: int = 100) -> List[Tuple[int, int, int]]:
+        """
+        Generate bright, distinct color palette for track visualization.
+        Uses HSV space with random hue and fixed high saturation/value.
+        """
+        np.random.seed(42)
+        colors = []
+        for i in range(num_colors):
+            hue = np.random.randint(0, 180)
+            saturation = 200 + np.random.randint(55)
+            value = 200 + np.random.randint(55)
+
+            hsv_color = np.uint8([[[hue, saturation, value]]])
+            bgr_color = cv2.cvtColor(hsv_color, cv2.COLOR_HSV2BGR)[0][0]
+            colors.append(tuple(map(int, bgr_color)))
+        return colors
+    
+    def _get_track_color(self, track_id: int) -> Tuple[int, int, int]:
+        """Get consistent color for track ID from palette."""
+        return self.color_palette[track_id % len(self.color_palette)]
+    
+    def _get_video_info(self, video_path: Path) -> Tuple[int, int, float, int]:
+        """
+        Get video metadata using ffmpeg.
+        
+        Returns:
+            Tuple of (width, height, fps, total_frames)
+        """
+        try:
+            probe = ffmpeg.probe(str(video_path))
+            video_stream = next((stream for stream in probe['streams'] 
+                               if stream['codec_type'] == 'video'), None)
+            
+            if video_stream is None:
+                raise ValueError(f"No video stream found in: {video_path}")
+            
+            width = int(video_stream['width'])
+            height = int(video_stream['height'])
+            
+            # Extract FPS
+            fps_str = video_stream.get('r_frame_rate', '30/1')
+            if '/' in fps_str:
+                num, den = map(int, fps_str.split('/'))
+                fps = num / den if den != 0 else 30.0
+            else:
+                fps = float(fps_str)
+            
+            # Get total frames
+            total_frames = int(video_stream.get('nb_frames', 0))
+            if total_frames == 0:
+                # Fallback: estimate from duration and fps
+                duration = float(video_stream.get('duration', 0))
+                total_frames = int(duration * fps) if duration > 0 else 0
+            
+            return width, height, fps, total_frames
+            
+        except Exception as e:
+            raise ValueError(f"Could not read video metadata {video_path}: {str(e)}")
+    
+    def _create_frame_iterator(self, source: Union[str, Path]) -> Iterator[Tuple[int, np.ndarray]]:
+        """
+        Create iterator that yields (frame_index, frame) tuples.
+        
+        Args:
+            source: Path to video file or directory with frame images
+            
+        Yields:
+            Tuple of (frame_index, frame_array)
+        """
+        source = Path(source)
+        
+        if source.is_file():
+            yield from self._iterate_video_frames(source)
+        elif source.is_dir():
+            yield from self._iterate_directory_frames(source)
+        else:
+            raise ValueError(f"Source must be a video file or directory, got: {source}")
+    
+    def _iterate_video_frames(self, video_path: Path) -> Iterator[Tuple[int, np.ndarray]]:
+        """Iterate through video frames using ffmpeg."""
+        width, height, fps, total_frames = self._get_video_info(video_path)
+        
+        # Store video info for later use
+        self.source_fps = fps
+        self.frame_size = (width, height)
+        
+        process = (
+            ffmpeg
+            .input(str(video_path))
+            .output('pipe:', format='rawvideo', pix_fmt='bgr24')
+            .run_async(pipe_stdout=True, pipe_stderr=True)
+        )
+        
+        try:
+            frame_size_bytes = width * height * 3
+            frame_idx = 0
+            
+            while True:
+                frame_data = process.stdout.read(frame_size_bytes)
+                if len(frame_data) != frame_size_bytes:
+                    break
+                
+                frame = np.frombuffer(frame_data, np.uint8).reshape([height, width, 3])
+                yield frame_idx, frame
+                frame_idx += 1
+                
+        finally:
+            process.stdout.close()
+            if process.stderr:
+                process.stderr.close()
+            process.wait()
+    
+    def _iterate_directory_frames(self, frames_dir: Path) -> Iterator[Tuple[int, np.ndarray]]:
+        """Iterate through image frames in directory."""
+        if not frames_dir.is_dir():
+            raise ValueError(f"Directory does not exist: {frames_dir}")
+        
+        # Support common image extensions
+        extensions = ['.jpg', '.jpeg', '.png', '.bmp', '.tiff']
+        image_files = []
+        for ext in extensions:
+            image_files.extend(frames_dir.glob(f'*{ext}'))
+            image_files.extend(frames_dir.glob(f'*{ext.upper()}'))
+        
+        image_files = sorted(image_files)
+        if not image_files:
+            raise ValueError(f"No image files found in directory: {frames_dir}")
+        
+        # Set fps from config for image sequences
+        self.source_fps = self.output_fps
+        
+        for frame_idx, img_path in enumerate(image_files):
+            frame = cv2.imread(str(img_path))
+            if frame is not None:
+                if frame_idx == 0:
+                    h, w = frame.shape[:2]
+                    self.frame_size = (w, h)
+                yield frame_idx, frame
+            else:
+                logger.warning(f"Could not read image: {img_path}")
+    
+    def _extract_tracks_from_annotation(self) -> None:
+        """
+        Extract tracking data from Supervisely VideoAnnotation.
+        
+        Populates self.tracks_by_frame with frame-indexed tracking data.
+        """
+        self.tracks_by_frame = defaultdict(list)
+        self.track_colors = {}
+        
+        # Map object keys to track info
+        objects = {}
+        for i, obj in enumerate(self.annotation.objects):
+            objects[obj.key] = (i, obj.obj_class.name)
+        
+        # Extract tracks from frames
+        for frame in self.annotation.frames:
+            frame_idx = frame.index
+            for figure in frame.figures:
+                if figure.geometry.geometry_name() != 'rectangle':
+                    continue
+                    
+                object_key = figure.parent_object.key
+                if object_key not in objects:
+                    continue
+                    
+                track_id, class_name = objects[object_key]
+                
+                # Extract bbox coordinates
+                rect = figure.geometry
+                bbox = (rect.left, rect.top, rect.right, rect.bottom)
+                
+                if track_id not in self.track_colors:
+                    if self.colorize_tracks:
+                        # auto-color override everything
+                        color = self._get_track_color(track_id)
+                    else:
+                        # try to use annotation color
+                        color = figure.video_object.obj_class.color
+                        if color:
+                            # convert rgb → bgr
+                            color = color[::-1]
+                        else:
+                            # fallback to auto-color if annotation missing
+                            color = self._get_track_color(track_id)
+
+                    self.track_colors[track_id] = color
+
+                
+                self.tracks_by_frame[frame_idx].append((track_id, bbox, class_name))
+        
+        logger.info(f"Extracted tracks from {len(self.tracks_by_frame)} frames")
+        
+    def _draw_detection(self, img: np.ndarray, track_id: int, bbox: Tuple[int, int, int, int], 
+                    class_name: str) -> Optional[Tuple[int, int]]:
+        """
+        Draw single detection with track ID and class label.
+        Returns the center point of the bbox for trajectory drawing.
+        """
+        x1, y1, x2, y2 = map(int, bbox)
+
+        if x2 <= x1 or y2 <= y1:
+            return None
+        
+        color = self.track_colors[track_id]
+
+        # Draw bounding box
+        cv2.rectangle(img, (x1, y1), (x2, y2), color, self.box_thickness)
+
+        # Draw label if enabled
+        if self.show_labels:
+            label = f"ID:{track_id}"
+            if self.show_classes:
+                label += f" ({class_name})"
+
+            label_y = y1 - 10 if y1 > 30 else y2 + 25
+            (text_w, text_h), _ = cv2.getTextSize(
+                label, cv2.FONT_HERSHEY_SIMPLEX, self.text_scale, self.text_thickness
+            )
+
+            cv2.rectangle(img, (x1, label_y - text_h - 5), 
+                        (x1 + text_w, label_y + 5), color, -1)
+            cv2.putText(img, label, (x1, label_y), 
+                    cv2.FONT_HERSHEY_SIMPLEX, self.text_scale, 
+                    (255, 255, 255), self.text_thickness, cv2.LINE_AA)
+
+        # Return center point for trajectory
+        return (x1 + x2) // 2, (y1 + y2) // 2
+
+
+    def _draw_trajectories(self, img: np.ndarray) -> None:
+        """Draw trajectory lines for all tracks, filtering out big jumps."""
+        if not self.show_trajectories:
+            return
+
+        max_jump = 200  
+
+        for track_id, centers in self.track_centers.items():
+            if len(centers) < 2:
+                continue
+
+            color = self.track_colors[track_id]
+            points = centers[-self.trajectory_length:]
+
+            for i in range(1, len(points)):
+                p1, p2 = points[i - 1], points[i]
+                if p1 is None or p2 is None:
+                    continue
+              
+                if np.hypot(p2[0] - p1[0], p2[1] - p1[1]) > max_jump:
+                    continue
+                cv2.line(img, p1, p2, color, 2)
+                cv2.circle(img, p1, 3, color, -1)
+
+        
+    def _process_single_frame(self, frame: np.ndarray, frame_idx: int) -> np.ndarray:
+        """
+        Process single frame: add annotations and return processed frame.
+        
+        Args:
+            frame: Input frame
+            frame_idx: Frame index
+            
+        Returns:
+            Annotated frame
+        """
+        img = frame.copy()
+        active_ids = set()
+        # Draw detections for current frame
+        if frame_idx in self.tracks_by_frame:
+            for track_id, bbox, class_name in self.tracks_by_frame[frame_idx]:
+                center = self._draw_detection(img, track_id, bbox, class_name)
+                self.track_centers[track_id].append(center)
+                if len(self.track_centers[track_id]) > self.trajectory_length:
+                    self.track_centers[track_id].pop(0)
+                active_ids.add(track_id)
+        
+        for tid in self.track_centers.keys():
+            if tid not in active_ids:
+                self.track_centers[tid].append(None)
+                if len(self.track_centers[tid]) > self.trajectory_length:
+                    self.track_centers[tid].pop(0)
+                
+        # Draw trajectories
+        self._draw_trajectories(img)
+        
+        # Add frame number if requested
+        if self.show_frame_number:
+            cv2.putText(img, f"Frame: {frame_idx + 1}", (10, 30),
+                       cv2.FONT_HERSHEY_SIMPLEX, 0.7, (255, 255, 255), 2, cv2.LINE_AA)
+        
+        return img
+    
+    def _save_processed_frame(self, frame: np.ndarray, frame_idx: int) -> str:
+        """
+        Save processed frame to temporary directory.
+        
+        Args:
+            frame: Processed frame
+            frame_idx: Frame index
+            
+        Returns:
+            Path to saved frame
+        """
+        frame_path = self._temp_dir / f"frame_{frame_idx:08d}.jpg"
+        cv2.imwrite(str(frame_path), frame, [cv2.IMWRITE_JPEG_QUALITY, 95])
+        return str(frame_path)
+    
+    def _create_video_from_frames(self, output_path: Union[str, Path]) -> None:
+        """
+        Create final video from processed frames using ffmpeg.
+        
+        Args:
+            output_path: Path for output video
+        """
+        output_path = Path(output_path)
+        output_path.parent.mkdir(parents=True, exist_ok=True)
+        
+        # Create video from frame sequence
+        input_pattern = str(self._temp_dir / "frame_%08d.jpg")
+        
+        try:
+            (
+                ffmpeg
+                .input(input_pattern, pattern_type='sequence', framerate=self.source_fps)
+                .output(str(output_path), vcodec='libx264', pix_fmt='yuv420p', crf=18)
+                .overwrite_output()
+                .run(capture_stdout=True, capture_stderr=True)
+            )
+            logger.info(f"Video saved to {output_path}")
+            
+        except ffmpeg.Error as e:
+            error_msg = e.stderr.decode() if e.stderr else "Unknown ffmpeg error"
+            raise ValueError(f"Failed to create video: {error_msg}")
+    
+    def _cleanup_temp_directory(self) -> None:
+        """Clean up temporary directory and all its contents."""
+        if self._temp_dir and self._temp_dir.exists():
+            shutil.rmtree(self._temp_dir)
+            self._temp_dir = None
+    
+    def visualize_video_annotation(self, annotation: VideoAnnotation, 
+                                  source: Union[str, Path], 
+                                  output_path: Union[str, Path]) -> None:
+        """
+        Visualize tracking annotations on video using streaming approach.
+        
+        Args:
+            annotation: Supervisely VideoAnnotation object with tracking data
+            source: Path to video file or directory containing frame images
+            output_path: Path for output video file
+            
+        Raises:
+            TypeError: If annotation is not VideoAnnotation
+            ValueError: If source is invalid or annotation is empty
+        """
+        if not isinstance(annotation, VideoAnnotation):
+            raise TypeError(f"Annotation must be VideoAnnotation, got {type(annotation)}")
+        
+        # Store annotation
+        self.annotation = annotation
+        
+        # Create temporary directory for processed frames
+        self._temp_dir = Path(tempfile.mkdtemp(prefix="video_viz_"))
+        
+        try:
+            # Extract tracking data
+            self._extract_tracks_from_annotation()
+            
+            if not self.tracks_by_frame:
+                logger.warning("No tracking data found in annotation")
+            
+            # Reset trajectory tracking
+            self.track_centers = defaultdict(list)
+            
+            # Process frames one by one
+            frame_count = 0
+            for frame_idx, frame in self._create_frame_iterator(source):
+                # Process frame
+                processed_frame = self._process_single_frame(frame, frame_idx)
+                
+                # Save processed frame
+                self._save_processed_frame(processed_frame, frame_idx)
+                
+                frame_count += 1
+                
+                # Progress logging
+                if frame_count % 100 == 0:
+                    logger.info(f"Processed {frame_count} frames")
+            
+            logger.info(f"Finished processing {frame_count} frames")
+            
+            # Create final video from saved frames
+            self._create_video_from_frames(output_path)
+            
+        finally:
+            # Always cleanup temporary files
+            self._cleanup_temp_directory()
+
+    def __del__(self):
+        """Cleanup temporary directory on object destruction."""
+        self._cleanup_temp_directory()
+        
+
+def visualize(
+    predictions: Union[VideoAnnotation, List[Prediction]], 
+    source: Union[str, Path], 
+    output_path: Union[str, Path],
+    show_labels: bool = True,
+    show_classes: bool = True,
+    show_trajectories: bool = True,
+    box_thickness: int = 2,
+    colorize_tracks: bool = True,
+    **kwargs
+) -> None:
+    """
+    Visualize tracking results from either VideoAnnotation or list of Prediction.
+
+    Args:
+        predictions (supervisely.VideoAnnotation | List[Prediction]): Tracking data to render; either a Supervisely VideoAnnotation or a list of Prediction objects.
+        source (str | Path): Path to an input video file or a directory of sequential frames (e.g., frame_000001.jpg).
+        output_path (str | Path): Path to the output video file to be created.
+        show_labels (bool, default=True): Draw per-object labels (track IDs).
+        show_classes (bool, default=True): Draw class names for each object.
+        show_trajectories (bool, default=True): Render object trajectories across frames.
+        box_thickness (int, default=2): Bounding-box line thickness in pixels.
+        colorize_tracks (bool, default=True): if True, ignore colors from project meta and generate new colors for each tracked object; if False, try to use colors from project meta when possible.    
+        """
+    visualizer = TrackingVisualizer(
+        show_labels=show_labels, 
+        show_classes=show_classes, 
+        show_trajectories=show_trajectories,
+        box_thickness=box_thickness,
+        colorize_tracks=colorize_tracks,
+        **kwargs
+    )
+
+    if isinstance(predictions, VideoAnnotation):
+        visualizer.visualize_video_annotation(predictions, source, output_path)
+    elif isinstance(predictions, list):
+        predictions = predictions_to_video_annotation(predictions)
+        visualizer.visualize_video_annotation(predictions, source, output_path)
+    else:
+        raise TypeError(f"Predictions must be VideoAnnotation or list of Prediction, got {type(predictions)}")
+

supervisely/template/experiment/experiment.html.jinja

@@ -174,6 +174,7 @@ can pass them in the terminal before running the script:
     model = api.nn.deploy(
         model="{{ experiment.paths.artifacts_dir.path }}/checkpoints/{{ experiment.training.checkpoints.pytorch.name }}",
         device="cuda:0", # or "cpu"
+        workspace_id={{ experiment.project.workspace_id }}
     )
 
     # 3. Predict
@@ -321,49 +322,41 @@ API](https://docs.supervisely.com/neural-networks/overview-1/prediction-api){:ta
 
 ## Tracking Objects in Video
 
-You can track objects in video using `boxmot` library.
-[BoxMot](https://github.com/mikel-brostrom/boxmot){:target="_blank"} is a
-third-party library that implements lightweight neural networks for tracking-by-detection task (when the tracking is
-performed on the objects predicted by a separate detector). For `boxmot` models you can use even CPU device.
+Supervisely now supports **tracking-by-detection** out of the box. We leverage a lightweight tracking algorithm (such as [BoT-SORT](https://github.com/NirAharon/BoT-SORT){:target="_blank"}) which identifies the unique objects across video frames and assigns IDs to them. This allows us to connect separate detections from different frames into a single track for each object.
 
-First, install [BoxMot](https://github.com/mikel-brostrom/boxmot){:target="_blank"}:
-
-```bash
-pip install boxmot
-```
-
-Supervisely SDK has the `track()` method from `supervisely.nn.tracking` which allows you to apply `boxmot` models
-together with a detector in a single line of code. This method takes two arguments: a `boxmot` tracker, and a
-`PredictionSession` of a detector. It returns a `sly.VideoAnnotation` with the tracked objects.
+To apply tracking via API, first, deploy your detection model or connect to it, and then use the `predict()` method with `tracking=True` parameter. You can also specify tracking configuration parameters by passing `tracking_config={...}` with your custom settings.
 
 ```python
 import supervisely as sly
-from supervisely.nn.tracking import track
-import boxmot
-from pathlib import Path
-
-# Deploy a detector
-detector = api.nn.deploy(
-    model="{{ experiment.model.framework }}/{{ experiment.model.name }}",
-    device="cuda:0", # Use GPU for detection
-)
 
-# Load BoxMot tracker
-tracker = boxmot.BotSort(
-    reid_weights=Path('osnet_x0_25_msmt17.pt'),
-    device="cpu", # Use CPU for tracking
+api = sly.Api()
+
+# Deploy your model
+model = api.nn.deploy(
+    model="{{ experiment.paths.artifacts_dir.path }}/checkpoints/{{ experiment.training.checkpoints.pytorch.name }}",
+    device="cuda",
+    workspace_id={{ experiment.project.workspace_id }},
 )
 
-# Track objects in a single line
-video_ann: sly.VideoAnnotation = track(
-    video_id=42,
-    detector=detector,
-    tracker=tracker,
+# Apply tracking
+predictions = model.predict(
+    video_id=YOUR_VIDEO_ID,  # Video ID in Supervisely
+    tracking=True,
+    tracking_config={
+        "tracker": "botsort",  # botsort is a powerful tracking algorithm used by default
+        # You can pass other tracking parameters here, see the docs for details
+    }
 )
+
+# Processing results
+for pred in predictions:
+    frame_index = pred.frame_index
+    annotation = pred.annotation
+    track_ids = pred.track_ids
+    print(f"Frame {frame_index}: {len(track_ids)} tracks")
 ```
 
-> For more information, see the section [Tracking Objects in
-Video](https://docs.supervisely.com/neural-networks/overview-1/prediction-api#tracking-objects-in-video){:target="_blank"}.
+> You can also apply trackers in your own code or applications. Read more about this in the docs [Video Object Tracking](https://docs.supervisely.com/neural-networks/overview-1/video-object-tracking){:target="_blank"}.
 
 {% endif %}
 

{supervisely-6.73.427.dist-info → supervisely-6.73.429.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: supervisely
-Version: 6.73.427
+Version: 6.73.429
 Summary: Supervisely Python SDK.
 Home-page: https://github.com/supervisely/supervisely
 Author: Supervisely