Cooked I Am

.vscode/settings.json

@@ -5,5 +5,5 @@
     "**/*.rpymc": true,
     "**/cache/": true
   },
-  "pylint.args": ["--generated-members", "cv2.*"],
+  "pylint.args": ["--generated-members=cv2.*", "--extension-pkg-whitelist=cv2"],
 }

main.py

@@ -8,6 +8,8 @@ from pipeline.core import SmokingDetectionPipeline
 from pipeline.models import YoloCigaretteDetector, YoloClassifier, YoloPersonDetector
 from pipeline.sinks import (
     FileAlertStore,
+    MetricsSink,
+    ScreenshotSink,
     SmartEventSink,
     VideoFileStore,
     WindowDisplaySink,
@@ -135,8 +137,10 @@ def main():
     sinks = []
     video_sink = None
     smart_sink = None
-    alert_store = None
+
-    display_sink = None
+    # Metrics sink (always active - provides performance summary)
+    metrics_sink = MetricsSink(log_interval=100)
+    sinks.append(metrics_sink)
 
     if not args.no_save:
         # Initialize Output Sink
@@ -148,6 +152,10 @@ def main():
         smart_sink = SmartEventSink("events", fps=source.fps)
         sinks.append(smart_sink)
 
+        # Screenshot sink (saves evidence JPGs)
+        screenshot_sink = ScreenshotSink("screenshots")
+        sinks.append(screenshot_sink)
+
     # Initialize Alert Store (always active)
     alert_store = FileAlertStore("alerts.jsonl")
     sinks.append(alert_store)

pipeline/core.py

@@ -5,8 +5,9 @@ from collections import deque
 from concurrent.futures import Future
 from typing import List, Optional, Tuple
 
+from pipeline.ops import nms_by_class
 from pipeline.protocols import Model, Sink, StreamSource, Verifier
-from pipeline.types import BoundingBox, Detection, Frame, Ticket, nms_by_class
+from pipeline.types import BoundingBox, Detection, Frame, Ticket
 
 logger = logging.getLogger(__name__)
 

pipeline/models/__init__.py

@@ -0,0 +1,15 @@
+"""Model wrappers for the pipeline."""
+
+from pipeline.models.yolo import (
+    DEVICE,
+    YoloCigaretteDetector,
+    YoloClassifier,
+    YoloPersonDetector,
+)
+
+__all__ = [
+    "DEVICE",
+    "YoloPersonDetector",
+    "YoloCigaretteDetector",
+    "YoloClassifier",
+]

pipeline/models/yolo.py

@@ -1,3 +1,5 @@
+"""YOLO model wrappers for detection and classification."""
+
 import logging
 from typing import Any, List
 
@@ -18,12 +20,13 @@ else:
 
 
 class YoloPersonDetector(Model):
-    def __init__(self, model_path: str = "yolo11n.pt"):  # Using small model for speed
+    """Detects people using YOLO."""
+
+    def __init__(self, model_path: str = "yolo11n.pt"):
         logger.info("Loading Person Detector: %s", model_path)
-        self.model = YOLO(model_path)  # Load model
+        self.model = YOLO(model_path)
 
     def predict(self, input_data: Any) -> List[Detection]:
-        # input_data: numpy array (H, W, C)
         results = self.model(
             input_data, classes=[0], verbose=False, device=DEVICE
         )  # 0 is person
@@ -45,13 +48,14 @@ class YoloPersonDetector(Model):
 
 
 class YoloCigaretteDetector(Model):
-    def __init__(self, model_path: str):
+    """Detects cigarettes using custom YOLO model."""
+
+    def __init__(self, model_path: str, confidence_threshold: float = 0.2):
         logger.info("Loading Cigarette Detector: %s", model_path)
         self.model = YOLO(model_path)
+        self.conf_threshold = confidence_threshold
 
     def predict(self, input_data: Any) -> List[Detection]:
-        # This model should be trained on cigarette class
-        # Assuming class 0 is cigarette for this custom model
         results = self.model(input_data, verbose=False, device=DEVICE)
 
         detections = []
@@ -60,8 +64,7 @@ class YoloCigaretteDetector(Model):
             for box in boxes:
                 x1, y1, x2, y2 = map(int, box.xyxy[0])
                 conf = float(box.conf[0])
-                # Maybe filter by conf here if needed
+                if conf > self.conf_threshold:
-                if conf > 0.2:
                     detections.append(
                         Detection(
                             box=BoundingBox(x1, y1, x2, y2, conf),
@@ -73,18 +76,18 @@ class YoloCigaretteDetector(Model):
 
 
 class YoloClassifier(Verifier):
+    """Classifies smoking using YOLO classification model."""
+
     def __init__(self, model_path: str):
-        logger.info("Loading Smoking Classifier: %s",  model_path)
+        logger.info("Loading Smoking Classifier: %s", model_path)
         self.model = YOLO(model_path)
 
     def verify(self, input_data: Any) -> float:
-        # input_data: numpy array (H, W, C) - Usually a crop
         if input_data.size == 0:
             return 0.0
 
         results = self.model(input_data, verbose=False, device=DEVICE)
 
-        # Check if we got any results
         if not results:
             return 0.0
 
@@ -92,16 +95,11 @@ class YoloClassifier(Verifier):
         if probs is None:
             return 0.0
 
-        # User specified Class 0 is "smoke"
+        # Class 0 is "smoke" - return its probability
-        # We return the probability of class 0 specifically.
-        # Ultralytics probs.data is a tensor of scores for each class.
         try:
             return float(probs.data[0])
         except (AttributeError, IndexError):
-            # Fallback if structure is unexpected, though data[0] is standard for classification
+            logger.warning("Could not retrieve Class 0 probability. Checking top1.")
-            logger.warning(
-                "Could not retrieve Class 0 probability. Defaulting to top1conf if class 0."
-            )
             if hasattr(probs, "top1") and probs.top1 == 0:
                 return float(probs.top1conf)
             return 0.0

pipeline/ops/__init__.py

@@ -0,0 +1,41 @@
+"""Operations and utility functions."""
+
+from pipeline.ops.drawing import (
+    draw_alert_banner,
+    draw_box,
+    draw_detections,
+    draw_fps,
+    draw_frame_overlay,
+    draw_label,
+    draw_zone,
+    get_color,
+)
+from pipeline.ops.geometry import (
+    box_center,
+    box_in_zone,
+    expand_box,
+    point_in_box,
+    point_in_polygon,
+)
+from pipeline.ops.nms import nms, nms_by_class
+
+__all__ = [
+    # NMS
+    "nms",
+    "nms_by_class",
+    # Geometry
+    "expand_box",
+    "point_in_box",
+    "box_center",
+    "point_in_polygon",
+    "box_in_zone",
+    # Drawing
+    "get_color",
+    "draw_box",
+    "draw_label",
+    "draw_detections",
+    "draw_frame_overlay",
+    "draw_alert_banner",
+    "draw_zone",
+    "draw_fps",
+]

pipeline/ops/drawing.py

@@ -0,0 +1,117 @@
+"""Drawing utilities for visualization."""
+
+from typing import List, Tuple
+
+import cv2
+import numpy as np
+
+from pipeline.types import Detection
+
+# Color palette (BGR)
+COLORS = {
+    "person": (255, 0, 0),  # Blue
+    "candidate": (0, 255, 255),  # Yellow
+    "verified": (0, 0, 255),  # Red
+    "default": (255, 255, 255),  # White
+}
+
+
+def get_color(label: str) -> Tuple[int, int, int]:
+    """Get color for a detection label."""
+    for key, color in COLORS.items():
+        if key in label:
+            return color
+    return COLORS["default"]
+
+
+def draw_box(
+    image: np.ndarray,
+    box: Tuple[int, int, int, int],
+    color: Tuple[int, int, int],
+    thickness: int = 2,
+):
+    """Draw a rectangle on image."""
+    x1, y1, x2, y2 = box
+    cv2.rectangle(image, (x1, y1), (x2, y2), color, thickness)
+
+
+def draw_label(
+    image: np.ndarray,
+    text: str,
+    position: Tuple[int, int],
+    color: Tuple[int, int, int],
+    font_scale: float = 0.5,
+):
+    """Draw a label with background."""
+    x, y = position
+    (tw, _), _ = cv2.getTextSize(text, cv2.FONT_HERSHEY_SIMPLEX, font_scale, 1)
+    cv2.rectangle(image, (x, y - 20), (x + tw, y), color, -1)
+    cv2.putText(
+        image, text, (x, y - 5), cv2.FONT_HERSHEY_SIMPLEX, font_scale, (0, 0, 0), 1
+    )
+
+
+def draw_detections(image: np.ndarray, detections: List[Detection]):
+    """Draw all detections with boxes and labels."""
+    for det in detections:
+        color = get_color(det.label)
+        draw_box(image, (det.box.x1, det.box.y1, det.box.x2, det.box.y2), color)
+        draw_label(
+            image, f"{det.label} {det.score:.2f}", (det.box.x1, det.box.y1), color
+        )
+
+
+def draw_frame_overlay(image: np.ndarray, frame_id: int):
+    """Draw frame counter overlay."""
+    cv2.putText(
+        image,
+        f"Frame: {frame_id}",
+        (10, 30),
+        cv2.FONT_HERSHEY_SIMPLEX,
+        1.0,
+        (0, 255, 0),
+        2,
+    )
+
+
+def draw_alert_banner(image: np.ndarray, text: str = "SMOKING DETECTED"):
+    """Draw a red alert banner at the top."""
+    cv2.rectangle(image, (0, 0), (image.shape[1], 40), (0, 0, 255), -1)
+    cv2.putText(
+        image,
+        text,
+        (10, 30),
+        cv2.FONT_HERSHEY_SIMPLEX,
+        1,
+        (255, 255, 255),
+        2,
+    )
+
+
+def draw_zone(
+    image: np.ndarray,
+    zone: List[Tuple[int, int]],
+    color: Tuple[int, int, int] = (0, 255, 0),
+    alpha: float = 0.3,
+):
+    """Draw a semi-transparent polygon zone."""
+    overlay = image.copy()
+    pts = np.array(zone, np.int32).reshape((-1, 1, 2))
+    cv2.fillPoly(overlay, [pts], color)
+    cv2.addWeighted(overlay, alpha, image, 1 - alpha, 0, image)
+    cv2.polylines(image, [pts], True, color, 2)
+
+
+def draw_fps(image: np.ndarray, fps: float):
+    """Draw FPS counter in top-right corner."""
+    h, w = image.shape[:2]
+    text = f"FPS: {fps:.1f}"
+    cv2.putText(
+        image,
+        text,
+        (w - 120, 30),
+        cv2.FONT_HERSHEY_SIMPLEX,
+        0.7,
+        (0, 255, 0),
+        2,
+    )

pipeline/ops/geometry.py

@@ -0,0 +1,73 @@
+"""Geometry utilities for bounding box operations."""
+
+from typing import List, Tuple
+
+from pipeline.types import BoundingBox
+
+
+def expand_box(
+    box: BoundingBox,
+    margin: float,
+    image_size: Tuple[int, int],
+) -> Tuple[int, int, int, int]:
+    """
+    Expand a bounding box by a percentage margin, clamped to image bounds.
+
+    Args:
+        box: Original bounding box
+        margin: Expansion factor (0.2 = 20% on each side)
+        image_size: (width, height) of the image
+
+    Returns:
+        (x1, y1, x2, y2) of expanded box
+    """
+    w_img, h_img = image_size
+    bw = box.x2 - box.x1
+    bh = box.y2 - box.y1
+
+    x1 = max(0, int(box.x1 - bw * margin))
+    y1 = max(0, int(box.y1 - bh * margin))
+    x2 = min(w_img, int(box.x2 + bw * margin))
+    y2 = min(h_img, int(box.y2 + bh * margin))
+
+    return x1, y1, x2, y2
+
+
+def point_in_box(point: Tuple[int, int], box: BoundingBox) -> bool:
+    """Check if a point (x, y) is inside a bounding box."""
+    x, y = point
+    return box.x1 <= x <= box.x2 and box.y1 <= y <= box.y2
+
+
+def box_center(box: BoundingBox) -> Tuple[int, int]:
+    """Get center point of a bounding box."""
+    cx = (box.x1 + box.x2) // 2
+    cy = (box.y1 + box.y2) // 2
+    return cx, cy
+
+
+def point_in_polygon(point: Tuple[int, int], polygon: List[Tuple[int, int]]) -> bool:
+    """
+    Ray casting algorithm to check if point is inside polygon.
+    Useful for zone filtering (ROI).
+    """
+    x, y = point
+    n = len(polygon)
+    inside = False
+
+    j = n - 1
+    for i in range(n):
+        xi, yi = polygon[i]
+        xj, yj = polygon[j]
+
+        if ((yi > y) != (yj > y)) and (x < (xj - xi) * (y - yi) / (yj - yi) + xi):
+            inside = not inside
+        j = i
+
+    return inside
+
+
+def box_in_zone(box: BoundingBox, zone: List[Tuple[int, int]]) -> bool:
+    """Check if box center is inside a zone polygon."""
+    center = box_center(box)
+    return point_in_polygon(center, zone)

pipeline/ops/nms.py

@@ -0,0 +1,61 @@
+"""Non-Maximum Suppression algorithms for detection filtering."""
+
+from typing import List
+
+from pipeline.types import Detection
+
+
+def nms(detections: List[Detection], iou_threshold: float = 0.5) -> List[Detection]:
+    """
+    Non-Maximum Suppression: Remove overlapping boxes, keeping the highest confidence ones.
+
+    Args:
+        detections: List of Detection objects
+        iou_threshold: Boxes with IoU > threshold are suppressed
+
+    Returns:
+        Filtered list of detections
+    """
+    if not detections:
+        return []
+
+    # Sort by confidence (highest first)
+    sorted_dets = sorted(detections, key=lambda d: d.score, reverse=True)
+
+    keep = []
+    while sorted_dets:
+        # Take the best one
+        best = sorted_dets.pop(0)
+        keep.append(best)
+
+        # Filter out boxes that overlap too much with 'best'
+        remaining = []
+        for det in sorted_dets:
+            if best.box.iou(det.box) < iou_threshold:
+                remaining.append(det)
+        sorted_dets = remaining
+
+    return keep
+
+
+def nms_by_class(
+    detections: List[Detection], iou_threshold: float = 0.5
+) -> List[Detection]:
+    """
+    Run NMS separately for each class label.
+    This prevents boxes of different classes from suppressing each other.
+    """
+    if not detections:
+        return []
+
+    # Group by label
+    by_label: dict[str, List[Detection]] = {}
+    for det in detections:
+        by_label.setdefault(det.label, []).append(det)
+
+    # Run NMS per class
+    result = []
+    for dets in by_label.values():
+        result.extend(nms(dets, iou_threshold))
+
+    return result

pipeline/sinks/__init__.py

@@ -3,10 +3,14 @@
 from pipeline.sinks.alerts import FileAlertStore
 from pipeline.sinks.display import WindowDisplaySink
 from pipeline.sinks.events import SmartEventSink
+from pipeline.sinks.metrics import MetricsSink
+from pipeline.sinks.screenshots import ScreenshotSink
 from pipeline.sinks.video import VideoFileStore
 
 __all__ = [
     "FileAlertStore",
+    "MetricsSink",
+    "ScreenshotSink",
     "SmartEventSink",
     "VideoFileStore",
     "WindowDisplaySink",

pipeline/sinks/base.py

@@ -1,63 +0,0 @@
-"""Shared drawing utilities for sinks."""
-
-from typing import List
-
-import cv2
-import numpy as np
-
-from pipeline.types import Detection
-
-
-def draw_detections(image: np.ndarray, detections: List[Detection]):
-    """Draw detection boxes with labels and color coding."""
-    for det in detections:
-        x1, y1, x2, y2 = det.box.x1, det.box.y1, det.box.x2, det.box.y2
-        label = det.label
-        score = det.score
-
-        # Color coding
-        color = (255, 255, 255)  # White default
-        if "person" in label:
-            color = (255, 0, 0)  # Blue
-        elif "candidate" in label:
-            color = (0, 255, 255)  # Yellow
-        elif "verified" in label:
-            color = (0, 0, 255)  # Red (Confirmed Smoke)
-
-        # Draw Box
-        cv2.rectangle(image, (x1, y1), (x2, y2), color, 2)
-
-        # Draw Label
-        text = f"{label} {score:.2f}"
-        (tw, th), _ = cv2.getTextSize(text, cv2.FONT_HERSHEY_SIMPLEX, 0.5, 1)
-        cv2.rectangle(image, (x1, y1 - 20), (x1 + tw, y1), color, -1)
-        cv2.putText(
-            image, text, (x1, y1 - 5), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 0), 1
-        )
-
-
-def draw_frame_overlay(image: np.ndarray, frame_id: int):
-    """Draw frame counter overlay."""
-    cv2.putText(
-        image,
-        f"Frame: {frame_id}",
-        (10, 30),
-        cv2.FONT_HERSHEY_SIMPLEX,
-        1.0,
-        (0, 255, 0),
-        2,
-    )
-
-
-def draw_alert_banner(image: np.ndarray, text: str = "SMOKING DETECTED"):
-    """Draw a red alert banner at the top."""
-    cv2.rectangle(image, (0, 0), (image.shape[1], 40), (0, 0, 255), -1)
-    cv2.putText(
-        image,
-        text,
-        (10, 30),
-        cv2.FONT_HERSHEY_SIMPLEX,
-        1,
-        (255, 255, 255),
-        2,
-    )

pipeline/sinks/display.py

@@ -4,8 +4,8 @@ import logging
 
 import cv2
 
+from pipeline.ops import draw_detections, draw_frame_overlay
 from pipeline.protocols import Sink
-from pipeline.sinks.base import draw_detections, draw_frame_overlay
 from pipeline.types import Frame
 
 logger = logging.getLogger(__name__)

pipeline/sinks/metrics.py

@@ -0,0 +1,102 @@
+"""Metrics sink - tracks pipeline performance stats."""
+
+import logging
+import time
+from collections import defaultdict
+from typing import Dict
+
+from pipeline.protocols import Sink
+from pipeline.types import Frame
+
+logger = logging.getLogger(__name__)
+
+
+class MetricsSink(Sink):
+    """
+    Tracks pipeline performance metrics:
+    - FPS (frames per second)
+    - Detection counts by label
+    - Alert count
+    - Processing latency
+    """
+
+    def __init__(self, log_interval: int = 100):
+        self.log_interval = log_interval
+
+        # Timing
+        self.start_time = time.time()
+        self.frame_count = 0
+        self.last_log_time = self.start_time
+
+        # Detection stats
+        self.detection_counts: Dict[str, int] = defaultdict(int)
+        self.alert_count = 0
+
+        # Latency tracking
+        self.frame_timestamps: list[float] = []
+
+        logger.info("MetricsSink initialized. Logging every %d frames.", log_interval)
+
+    def handle(self, frame: Frame):
+        self.frame_count += 1
+        now = time.time()
+
+        # Track latency (time from capture to processing)
+        latency = now - frame.timestamp
+        self.frame_timestamps.append(latency)
+        # Keep only last 100 for rolling average
+        if len(self.frame_timestamps) > 100:
+            self.frame_timestamps.pop(0)
+
+        # Count detections by label
+        for det in frame.detections:
+            self.detection_counts[det.label] += 1
+
+        # Count alerts
+        if frame.ticket:
+            self.alert_count += 1
+
+        # Periodic logging
+        if self.frame_count % self.log_interval == 0:
+            self._log_stats(now)
+
+    def _log_stats(self, now: float):
+        elapsed = now - self.start_time
+        fps = self.frame_count / elapsed if elapsed > 0 else 0
+        avg_latency = (
+            sum(self.frame_timestamps) / len(self.frame_timestamps)
+            if self.frame_timestamps
+            else 0
+        )
+
+        logger.info(
+            "[METRICS] Frames: %d | FPS: %.1f | Latency: %.1fms | Alerts: %d",
+            self.frame_count,
+            fps,
+            avg_latency * 1000,
+            self.alert_count,
+        )
+
+    def close(self):
+        """Print final summary."""
+        elapsed = time.time() - self.start_time
+        fps = self.frame_count / elapsed if elapsed > 0 else 0
+        avg_latency = (
+            sum(self.frame_timestamps) / len(self.frame_timestamps)
+            if self.frame_timestamps
+            else 0
+        )
+
+        logger.info("=" * 50)
+        logger.info("PIPELINE METRICS SUMMARY")
+        logger.info("=" * 50)
+        logger.info("Total Frames:    %d", self.frame_count)
+        logger.info("Runtime:         %.1f seconds", elapsed)
+        logger.info("Average FPS:     %.1f", fps)
+        logger.info("Avg Latency:     %.1f ms", avg_latency * 1000)
+        logger.info("Total Alerts:    %d", self.alert_count)
+        logger.info("-" * 50)
+        logger.info("Detections by Label:")
+        for label, count in sorted(self.detection_counts.items()):
+            logger.info("  %-25s %d", label, count)
+        logger.info("=" * 50)

pipeline/sinks/screenshots.py

@@ -0,0 +1,53 @@
+"""Screenshot sink - saves evidence images on alerts."""
+
+import logging
+from pathlib import Path
+
+import cv2
+
+from pipeline.ops.drawing import draw_alert_banner, draw_detections
+from pipeline.protocols import Sink
+from pipeline.types import Frame
+
+logger = logging.getLogger(__name__)
+
+
+class ScreenshotSink(Sink):
+    """
+    Saves a screenshot when a ticket (alert) fires.
+    Useful for evidence/audit trails without full video recording.
+    """
+
+    def __init__(self, output_dir: str = "screenshots", annotate: bool = True):
+        self.output_dir = Path(output_dir)
+        self.output_dir.mkdir(parents=True, exist_ok=True)
+        self.annotate = annotate
+        self.count = 0
+
+        logger.info("ScreenshotSink initialized. Saving to %s", output_dir)
+
+    def handle(self, frame: Frame):
+        if not frame.ticket:
+            return
+
+        self.count += 1
+        ticket = frame.ticket
+
+        # Build filename with timestamp and confidence
+        filename = f"alert_{ticket.frame_id}_{ticket.confidence:.2f}.jpg"
+        filepath = self.output_dir / filename
+
+        # Prepare image
+        if self.annotate:
+            img = frame.data.copy()
+            draw_detections(img, frame.detections)
+            draw_alert_banner(img, f"SMOKING {ticket.confidence:.0%}")
+        else:
+            img = frame.data
+
+        # Save
+        cv2.imwrite(str(filepath), img)
+        logger.info("Saved screenshot: %s", filepath)
+
+    def close(self):
+        logger.info("ScreenshotSink closed. Total screenshots: %d", self.count)

pipeline/sinks/video.py

@@ -9,8 +9,8 @@ from typing import Optional
 
 import cv2
 
+from pipeline.ops import draw_detections, draw_frame_overlay
 from pipeline.protocols import Sink
-from pipeline.sinks.base import draw_detections, draw_frame_overlay
 from pipeline.types import Frame
 
 logger = logging.getLogger(__name__)

pipeline/sources.py

@@ -1,133 +0,0 @@
-import logging
-import threading
-import time
-from queue import Empty, Queue
-from typing import Iterator
-
-import cv2
-
-from pipeline.protocols import StreamSource
-from pipeline.types import Frame
-
-logger = logging.getLogger(__name__)
-
-
-class FileStreamSource(StreamSource):
-    def __init__(self, path: str):
-        self.path = path
-        self._fps = 30.0
-
-        # Probe for FPS if it's a single file
-        if "*" not in self.path:
-            try:
-                cap = cv2.VideoCapture(self.path)
-                if cap.isOpened():
-                    self._fps = cap.get(cv2.CAP_PROP_FPS) or 30.0
-                cap.release()
-            except Exception as e:
-                logger.warning("Could not probe FPS from %s: %s", path, e)
-
-    @property
-    def fps(self) -> float:
-        return self._fps
-
-    def stream(self) -> Iterator[Frame]:
-        # Handle glob patterns if path is a directory
-        if "*" in self.path:
-            import glob
-
-            files = glob.glob(self.path)
-            # Todo: handle image sequence
-            return
-
-        logger.info("Opening video file: %s", self.path)
-        cap = cv2.VideoCapture(self.path)
-        frame_id = 0
-        while cap.isOpened():
-            ret, frame = cap.read()
-            if not ret:
-                logger.info("End of video file reached.")
-                break
-
-            yield Frame(timestamp=time.time(), data=frame, frame_id=frame_id)
-            frame_id += 1
-        cap.release()
-
-    def stop(self):
-        """No-op for file source (stream ends naturally)."""
-        pass
-
-
-class ThreadedRTSPSource(StreamSource):
-    """
-    Reads from RTSP stream or webcam in a background thread to prevent buffer buildup
-    and allow main processing loop to run at its own pace without stalling the decoding.
-    """
-
-    def __init__(self, source: str | int, queue_size: int = 5):
-        self.source = source
-        self.queue: Queue[Frame] = Queue(maxsize=queue_size)
-        self._fps = 30.0
-
-        # Probe FPS (with timeout safety ideally, but simple here)
-        try:
-            cap = cv2.VideoCapture(self.source)
-            if cap.isOpened():
-                self._fps = cap.get(cv2.CAP_PROP_FPS) or 30.0
-            cap.release()
-        except Exception:
-            logger.warning("Could not probe FPS from source.")
-
-        self.stopped = False
-        self.thread = threading.Thread(target=self._update, daemon=True)
-        self._start()
-        logger.info("Initialized ThreadedRTSPSource for %s", source)
-
-    @property
-    def fps(self) -> float:
-        return self._fps
-
-    def _start(self):
-        self.thread.start()
-
-    def _update(self):
-        logger.info("Reader thread started for %s", self.source)
-        cap = cv2.VideoCapture(self.source)
-        frame_id = 0
-        while not self.stopped:
-            if not cap.isOpened():
-                # Try reconnecting
-                logger.warning("Stream not open, retrying...")
-                time.sleep(1)
-                cap = cv2.VideoCapture(self.source)
-                continue
-
-            ret, frame = cap.read()
-            if not ret:
-                # Stream ended or failed
-                logger.warning("Failed to read frame from RTSP, retrying...")
-                time.sleep(0.1)
-                continue
-
-            # If queue is full, remove the oldest frame to keep latest
-            if self.queue.full():
-                try:
-                    self.queue.get_nowait()
-                except Empty:
-                    pass
-
-            self.queue.put(Frame(timestamp=time.time(), data=frame, frame_id=frame_id))
-            frame_id += 1
-        cap.release()
-        logger.info("RTSP Thread stopped.")
-
-    def stream(self) -> Iterator[Frame]:
-        while not self.stopped:
-            try:
-                yield self.queue.get(timeout=1)
-            except Empty:
-                continue
-
-    def stop(self):
-        self.stopped = True
-        self.thread.join()

pipeline/sources/__init__.py

@@ -0,0 +1,13 @@
+"""Stream sources for the pipeline."""
+
+from pipeline.sources.file import FileStreamSource
+from pipeline.sources.stream import ThreadedStreamSource
+
+# Alias for backwards compatibility
+ThreadedRTSPSource = ThreadedStreamSource
+
+__all__ = [
+    "FileStreamSource",
+    "ThreadedStreamSource",
+    "ThreadedRTSPSource",  # Alias
+]

pipeline/sources/file.py

@@ -0,0 +1,63 @@
+"""File-based video source."""
+
+import glob
+import logging
+import time
+from typing import Iterator
+
+import cv2
+
+from pipeline.protocols import StreamSource
+from pipeline.types import Frame
+
+logger = logging.getLogger(__name__)
+
+
+class FileStreamSource(StreamSource):
+    """Reads frames from a video file."""
+
+    def __init__(self, path: str):
+        self.path = path
+        self._fps = 30.0
+
+        # Probe for FPS if it's a single file
+        if "*" not in self.path:
+            try:
+                cap = cv2.VideoCapture(self.path)
+                if cap.isOpened():
+                    self._fps = cap.get(cv2.CAP_PROP_FPS) or 30.0
+                cap.release()
+            except Exception as e:
+                logger.warning("Could not probe FPS from %s: %s", path, e)
+
+    @property
+    def fps(self) -> float:
+        return self._fps
+
+    def stream(self) -> Iterator[Frame]:
+        # Handle glob patterns for image sequences
+        if "*" in self.path:
+            files = sorted(glob.glob(self.path))
+            logger.info("Loading %d images from pattern: %s", len(files), self.path)
+            for frame_id, file in enumerate(files):
+                img = cv2.imread(file)
+                if img is not None:
+                    yield Frame(timestamp=time.time(), data=img, frame_id=frame_id)
+            return
+
+        logger.info("Opening video file: %s", self.path)
+        cap = cv2.VideoCapture(self.path)
+        frame_id = 0
+        while cap.isOpened():
+            ret, frame = cap.read()
+            if not ret:
+                logger.info("End of video file reached.")
+                break
+
+            yield Frame(timestamp=time.time(), data=frame, frame_id=frame_id)
+            frame_id += 1
+        cap.release()
+
+    def stop(self):
+        """No-op for file source (stream ends naturally)."""
+        pass

pipeline/sources/stream.py

@@ -0,0 +1,92 @@
+"""Threaded source for RTSP streams and webcams."""
+
+import logging
+import threading
+import time
+from queue import Empty, Queue
+from typing import Iterator
+
+import cv2
+
+from pipeline.protocols import StreamSource
+from pipeline.types import Frame
+
+logger = logging.getLogger(__name__)
+
+
+class ThreadedStreamSource(StreamSource):
+    """
+    Reads from RTSP stream or webcam in a background thread.
+
+    Prevents buffer buildup and allows main processing loop to run
+    at its own pace without stalling the decoding.
+    """
+
+    def __init__(self, source: str | int, queue_size: int = 5):
+        self.source = source
+        self.queue: Queue[Frame] = Queue(maxsize=queue_size)
+        self._fps = 30.0
+
+        # Probe FPS
+        try:
+            cap = cv2.VideoCapture(self.source)
+            if cap.isOpened():
+                self._fps = cap.get(cv2.CAP_PROP_FPS) or 30.0
+            cap.release()
+        except cv2.error: # pylint: disable=catching-non-exception
+            logger.warning("Could not probe FPS from source.")
+
+        self.stopped = False
+        self.thread = threading.Thread(target=self._reader_loop, daemon=True)
+        self.thread.start()
+        logger.info("Initialized ThreadedStreamSource for %s", source)
+
+    @property
+    def fps(self) -> float:
+        return self._fps
+
+    def _reader_loop(self):
+        """Background thread that continuously reads frames."""
+        logger.info("Reader thread started for %s", self.source)
+        cap = cv2.VideoCapture(self.source)
+        frame_id = 0
+
+        while not self.stopped:
+            if not cap.isOpened():
+                logger.warning("Stream not open, retrying...")
+                time.sleep(1)
+                cap = cv2.VideoCapture(self.source)
+                continue
+
+            ret, frame = cap.read()
+            if not ret:
+                logger.warning("Failed to read frame, retrying...")
+                time.sleep(0.1)
+                continue
+
+            # Drop oldest if queue full (keep latest frames)
+            if self.queue.full():
+                try:
+                    self.queue.get_nowait()
+                except Empty:
+                    pass
+
+            self.queue.put(Frame(timestamp=time.time(), data=frame, frame_id=frame_id))
+            frame_id += 1
+
+        cap.release()
+        logger.info("Reader thread stopped.")
+
+    def stream(self) -> Iterator[Frame]:
+        """Yield frames from the queue."""
+        while not self.stopped:
+            try:
+                yield self.queue.get(timeout=1)
+            except Empty:
+                continue
+
+    def stop(self):
+        """Stop the reader thread and release resources."""
+        self.stopped = True
+        if self.thread.is_alive():
+            self.thread.join(timeout=2.0)

pipeline/types.py

@@ -58,62 +58,6 @@ class Detection:
     metadata: dict = field(default_factory=dict)
 
 
-def nms(detections: List[Detection], iou_threshold: float = 0.5) -> List[Detection]:
-    """
-    Non-Maximum Suppression: Remove overlapping boxes, keeping the highest confidence ones.
-
-    Args:
-        detections: List of Detection objects
-        iou_threshold: Boxes with IoU > threshold are suppressed
-
-    Returns:
-        Filtered list of detections
-    """
-    if not detections:
-        return []
-
-    # Sort by confidence (highest first)
-    sorted_dets = sorted(detections, key=lambda d: d.score, reverse=True)
-
-    keep = []
-    while sorted_dets:
-        # Take the best one
-        best = sorted_dets.pop(0)
-        keep.append(best)
-
-        # Filter out boxes that overlap too much with 'best'
-        remaining = []
-        for det in sorted_dets:
-            if best.box.iou(det.box) < iou_threshold:
-                remaining.append(det)
-        sorted_dets = remaining
-
-    return keep
-
-
-def nms_by_class(
-    detections: List[Detection], iou_threshold: float = 0.5
-) -> List[Detection]:
-    """
-    Run NMS separately for each class label.
-    This prevents boxes of different classes from suppressing each other.
-    """
-    if not detections:
-        return []
-
-    # Group by label
-    by_label: dict[str, List[Detection]] = {}
-    for det in detections:
-        by_label.setdefault(det.label, []).append(det)
-
-    # Run NMS per class
-    result = []
-    for label, dets in by_label.items():
-        result.extend(nms(dets, iou_threshold))
-
-    return result
-
-
 @dataclass
 class Ticket:
     """Represents a potential alert event."""