remove status, add track score, v0.0.9

Author: wmuron

README.md

@@ -126,4 +126,4 @@ Feel free to tune the parameter of Q and R matrix builders to better fit your us
 ## References, papers, ideas and acknowledgements
 - https://github.com/rlabbe/Kalman-and-Bayesian-Filters-in-Python/
 - http://elvera.nue.tu-berlin.de/files/1517Bochinski2017.pdf
-- https://arxiv.org/abs/1602.00763
+- https://arxiv.org/abs/1602.00763

examples/2d_multi_object_tracking.py

@@ -31,11 +31,12 @@ def demo_tracking_visualization(num_steps: int = 1000, num_objects: int = 10):
         active_tracks = tracker.step(detections=detections)
 
         for track in active_tracks:
+            score = track.score if track.score is not None else -1
             img = draw_rectangle(img, track.box, color=(10, 10, 220), thickness=5)
-            img = draw_text(img, track.id, above_box=track.box)
+            img = draw_text(img, f'{track.id[:8]}... ({score:.2f})', above_box=track.box)
 
         for det in detections:
-            img = draw_rectangle(img, det.box, color=(10, 220, 20), thickness=2)
+            img = draw_rectangle(img, det.box, color=(10, 220, 20), thickness=1)
 
         cv2.imshow('preview', img)
         # stop the demo by pressing q

examples/webcam_face_tracking.py

@@ -3,7 +3,7 @@
 from urllib.request import urlretrieve
 
 import cv2
-from motpy import Detection, MultiObjectTracker, NpImage
+from motpy import Detection, MultiObjectTracker, NpImage, Box
 from motpy.core import setup_logger
 from motpy.detector import BaseObjectDetector
 from motpy.testing_viz import draw_detection, draw_track
@@ -15,7 +15,7 @@
 
 """
 
-logger = setup_logger(__name__, is_main=True)
+logger = setup_logger(__name__, 'DEBUG', is_main=True)
 
 
 WEIGHTS_URL = 'https://github.com/opencv/opencv_3rdparty/raw/dnn_samples_face_detector_20170830/res10_300x300_ssd_iter_140000.caffemodel'
@@ -30,7 +30,7 @@ def __init__(self,
                  weights_path: str = WEIGHTS_PATH,
                  config_url: str = CONFIG_URL,
                  config_path: str = CONFIG_PATH,
-                 conf_threshold: float = 0.5):
+                 conf_threshold: float = 0.5) -> None:
         super(FaceDetector, self).__init__()
 
         if not os.path.isfile(weights_path) or not os.path.isfile(config_path):
@@ -43,23 +43,23 @@ def __init__(self,
         # specify detector hparams
         self.conf_threshold = conf_threshold
 
-    def process_image(self, image: NpImage) -> Sequence[NpImage]:
+    def process_image(self, image: NpImage) -> Sequence[Detection]:
         blob = cv2.dnn.blobFromImage(image, 1.0, (300, 300), [104, 117, 123], False, False)
         self.net.setInput(blob)
         detections = self.net.forward()
 
         # convert output from OpenCV detector to tracker expected format [xmin, ymin, xmax, ymax]
-        bboxes = []
+        out_detections = []
         for i in range(detections.shape[2]):
             confidence = detections[0, 0, i, 2]
             if confidence > self.conf_threshold:
                 xmin = int(detections[0, 0, i, 3] * image.shape[1])
                 ymin = int(detections[0, 0, i, 4] * image.shape[0])
                 xmax = int(detections[0, 0, i, 5] * image.shape[1])
                 ymax = int(detections[0, 0, i, 6] * image.shape[0])
-                bboxes.append([xmin, ymin, xmax, ymax])
-
-        return bboxes
+                out_detections.append(Detection(box=[xmin, ymin, xmax, ymax], score=confidence))
+            
+        return out_detections
 
 
 def run():
@@ -84,8 +84,7 @@ def run():
         frame = cv2.resize(frame, dsize=None, fx=0.5, fy=0.5)
 
         # run face detector on current frame
-        bboxes = face_detector.process_image(frame)
-        detections = [Detection(box=bbox) for bbox in bboxes]
+        detections = face_detector.process_image(frame)
         logger.debug(f'detections: {detections}')
 
         tracker.step(detections)

motpy/core.py

@@ -15,14 +15,13 @@
 Vector = np.ndarray
 
 # Track is meant as an output from the object tracker
-Track = collections.namedtuple('Track', 'id box')
+Track = collections.namedtuple('Track', 'id box score')
 
 # numpy/opencv image alias
 NpImage = np.ndarray
 
 
 class Detection:
-    # Detection is to be an input the the tracker
     def __init__(
             self,
             box: Box,
@@ -33,10 +32,7 @@ def __init__(
         self.feature = feature
 
     def __repr__(self):
-        fmt = "(detection: box=%s, score=%s, feature=%s)"
-        return fmt % (str(self.box),
-                      str(self.score) or 'none',
-                      str(self.feature) or 'none')
+        return f'Detection(box={self.box}, score={self.score:.5f}, feature={self.feature})'
 
 
 """ utils """
@@ -51,17 +47,17 @@ def setup_logger(name: str,
     if level is None:
         level = os.getenv(envvar_name)
         if level is None:
-            print(f'[{name}] fallback to INFO log_level; set {envvar_name} envvar to override')
             level = 'INFO'
         else:
             print(f'[{name}] envvar {envvar_name} sets log level to {level}')
 
     level_val = logging.getLevelName(level)
 
-    if is_main:
-        logging.basicConfig(stream=sys.stdout, level=level_val, format=LOG_FORMAT)
-
     logger = logging.getLogger(name)
     logger.setLevel(level_val)
+    logger.addHandler(logging.NullHandler())
+
+    if is_main:
+        logging.basicConfig(stream=sys.stdout, level=level_val, format=LOG_FORMAT)
 
     return logger

motpy/detector.py

@@ -1,6 +1,5 @@
 from typing import Sequence
 
-import numpy as np
 from motpy.core import NpImage
 from motpy.tracker import Detection
 

motpy/metrics.py

@@ -10,13 +10,13 @@ def calculate_iou(bboxes1, bboxes2, dim: int = 2):
     coords_b1 = np.split(bboxes1, 2 * dim, axis=1)
     coords_b2 = np.split(bboxes2, 2 * dim, axis=1)
 
-    coords_tl, coords_br = dict(), dict()  # tl and br points
+    coords = np.zeros(shape=(2, dim, bboxes1.shape[0], bboxes2.shape[0]))
     val_inter, val_b1, val_b2 = 1.0, 1.0, 1.0
     for d in range(dim):
-        coords_tl[d] = np.maximum(coords_b1[d], np.transpose(coords_b2[d]))
-        coords_br[d] = np.minimum(coords_b1[d + dim], np.transpose(coords_b2[d + dim]))
+        coords[0, d] = np.maximum(coords_b1[d], np.transpose(coords_b2[d]))  # top-left
+        coords[1, d] = np.minimum(coords_b1[d + dim], np.transpose(coords_b2[d + dim]))  # bottom-right
 
-        val_inter *= np.maximum(coords_br[d] - coords_tl[d], 0)
+        val_inter *= np.maximum(coords[1, d] - coords[0, d], 0)
         val_b1 *= coords_b1[d + dim] - coords_b1[d]
         val_b2 *= coords_b2[d + dim] - coords_b2[d]
 

motpy/model.py

@@ -1,13 +1,10 @@
-
 from enum import Enum
 
 import numpy as np
 from filterpy.common import Q_discrete_white_noise
-from filterpy.kalman import KalmanFilter
-from motpy.core import Box, Detection, Track, Vector
-from motpy.metrics import angular_similarity, calculate_iou
 from scipy.linalg import block_diag
 
+from motpy.core import Box, Vector
 
 """ The list of model presets below is not complete, more reasonable
 options will be added in the future """
@@ -140,7 +137,7 @@ def box_to_z(self, box: Box) -> Vector:
 
     def box_to_x(self, box: Box) -> Vector:
         """ box is expected to be in [xmin, ymin, zmin, ..., xmax, ymax, zmax, ...] format
-        for 2d1ord+2d0ord case returns np.array([cx, 0, 0, cy, 0, 0, w, h]) """
+        for 2d-1ord+2d-0ord case returns np.array([cx, 0, 0, cy, 0, 0, w, h]) """
         x = np.zeros((self.state_length,))
         x[self.z_in_x_idxs] = self.box_to_z(box)
         return x

motpy/testing.py

@@ -1,8 +1,6 @@
 import math
 import random
 
-import numpy as np
-
 from motpy.core import Detection
 
 CANVAS_SIZE = 1000
@@ -73,11 +71,11 @@ def detections(self, step: int):
             self.disappear_steps -= 1
 
         # wrap boxes and features as detections
-        det_gt = Detection(box=box_gt, score=1.0, feature=self.color)
+        det_gt = Detection(box=box_gt, score=1., feature=self.color)
 
         feature_pred = [random.gauss(0, 5) + v for v in self.color]
         det_pred = Detection(box=box_pred,
-                             score=random.random() / 2 + 0.5,
+                             score=random.uniform(0.5, 1.),
                              feature=feature_pred)
 
         return det_gt, det_pred