warden-vision 0.1.2

Reusable computer vision building blocks for detection, tracking, and annotation, in the spirit of Roboflow's Python `supervision` library.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150

"""Benchmark counterpart to examples/bench_vs_supervision.rs in warden-vision.

Same three scenarios, same sizes, so numbers are directly comparable:
1. NMS over 10,000 random boxes.
2. ByteTrack over 500 frames x 30 detections.
3. Box + label annotation over 500 frames x 50 boxes on a 1920x1080 image.
"""

import time
import warnings

import numpy as np
import supervision as sv

warnings.filterwarnings("ignore", category=FutureWarning)


class Rng:
    """Same xorshift PRNG as the Rust side, so both benchmarks draw identical data."""

    def __init__(self, seed):
        self.state = seed

    def next_f32(self):
        x = self.state
        x ^= (x << 13) & 0xFFFFFFFFFFFFFFFF
        x ^= x >> 7
        x ^= (x << 17) & 0xFFFFFFFFFFFFFFFF
        self.state = x
        return (x >> 11) / float(1 << 53)


def bench_nms():
    rng = Rng(42)
    n = 10_000
    xyxy = np.zeros((n, 4), dtype=np.float32)
    confidence = np.zeros(n, dtype=np.float32)
    class_id = np.zeros(n, dtype=int)
    for i in range(n):
        x = rng.next_f32() * 1900.0
        y = rng.next_f32() * 1000.0
        w = 20.0 + rng.next_f32() * 80.0
        h = 20.0 + rng.next_f32() * 80.0
        xyxy[i] = [x, y, x + w, y + h]
        confidence[i] = rng.next_f32()
        class_id[i] = i % 10

    detections = sv.Detections(xyxy=xyxy, confidence=confidence, class_id=class_id)

    iters = 20
    start = time.perf_counter()
    kept = 0
    for _ in range(iters):
        kept = len(detections.with_nms(threshold=0.5, class_agnostic=False))
    elapsed = time.perf_counter() - start
    print(
        f"[nms]        n={n:5} iters={iters:3} kept={kept:5} "
        f"total={elapsed * 1000:8.2f}ms avg={elapsed / iters * 1000:8.2f}ms"
    )


def bench_bytetrack():
    rng = Rng(7)
    frames = 500
    per_frame = 30

    all_frames = []
    for f in range(frames):
        xyxy = np.zeros((per_frame, 4), dtype=np.float32)
        confidence = np.zeros(per_frame, dtype=np.float32)
        class_id = np.zeros(per_frame, dtype=int)
        for i in range(per_frame):
            base_x = i * 60.0 + f * 0.5
            base_y = i * 15.0
            jitter = rng.next_f32() * 4.0 - 2.0
            xyxy[i] = [
                base_x + jitter,
                base_y + jitter,
                base_x + 40.0 + jitter,
                base_y + 40.0 + jitter,
            ]
            confidence[i] = 0.5 + rng.next_f32() * 0.5
            class_id[i] = i % 5
        all_frames.append(sv.Detections(xyxy=xyxy, confidence=confidence, class_id=class_id))

    tracker = sv.ByteTrack(
        track_activation_threshold=0.25,
        lost_track_buffer=30,
        minimum_matching_threshold=0.8,
    )

    start = time.perf_counter()
    total_tracked = 0
    for frame in all_frames:
        total_tracked += len(tracker.update_with_detections(frame))
    elapsed = time.perf_counter() - start
    print(
        f"[bytetrack]  frames={frames:4} per_frame={per_frame:3} tracked={total_tracked:6} "
        f"total={elapsed * 1000:8.2f}ms fps={frames / elapsed:8.1f}"
    )


def bench_annotate():
    rng = Rng(99)
    frames = 500
    per_frame = 50
    w, h = 1920, 1080

    all_frames = []
    for _ in range(frames):
        xyxy = np.zeros((per_frame, 4), dtype=np.float32)
        confidence = np.zeros(per_frame, dtype=np.float32)
        class_id = np.zeros(per_frame, dtype=int)
        tracker_id = np.zeros(per_frame, dtype=int)
        for i in range(per_frame):
            x = rng.next_f32() * (w - 100.0)
            y = rng.next_f32() * (h - 100.0)
            xyxy[i] = [x, y, x + 80.0, y + 60.0]
            confidence[i] = 0.9
            class_id[i] = i % 10
            tracker_id[i] = i
        all_frames.append(
            sv.Detections(
                xyxy=xyxy,
                confidence=confidence,
                class_id=class_id,
                tracker_id=tracker_id,
            )
        )

    box_annotator = sv.BoxAnnotator(thickness=2)
    label_annotator = sv.LabelAnnotator(text_scale=0.5, text_thickness=1, text_padding=4)

    start = time.perf_counter()
    for dets in all_frames:
        frame = np.zeros((h, w, 3), dtype=np.uint8)
        frame = box_annotator.annotate(frame, dets)
        frame = label_annotator.annotate(frame, dets)
    elapsed = time.perf_counter() - start
    print(
        f"[annotate]   frames={frames:4} per_frame={per_frame:3} {w}x{h} "
        f"total={elapsed * 1000:8.2f}ms fps={frames / elapsed:8.1f}"
    )


if __name__ == "__main__":
    print(f"supervision {sv.__version__} benchmark")
    bench_nms()
    bench_bytetrack()
    bench_annotate()