File size: 7,435 Bytes
9422383 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 | #!/usr/bin/env python3
"""Run YOLO26n detection on prerecorded video."""
from __future__ import annotations
import argparse
import hashlib
import json
from dataclasses import dataclass
from pathlib import Path
from time import perf_counter
import cv2
import numpy as np
from arm_yolo26n import DetectionResult, Yolo26Detector
@dataclass(frozen=True, slots=True)
class _Arguments:
runtime: Path
model: Path
video: Path
output: Path
frame_count: int
@dataclass(frozen=True, slots=True)
class _PreloadedVideo:
frames: tuple[np.ndarray, ...]
source_width: int
source_height: int
source_fps: float
decode_resize_ms: float
def _digest(path: Path) -> str:
value = hashlib.sha256()
with path.open("rb") as stream:
for block in iter(lambda: stream.read(1024 * 1024), b""):
value.update(block)
return value.hexdigest()
def _parse_args() -> _Arguments:
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("--runtime", type=Path, required=True)
parser.add_argument("--model", type=Path, required=True)
parser.add_argument("--video", type=Path, required=True)
parser.add_argument("--output", type=Path, required=True)
parser.add_argument("--frames", type=int, default=200)
args = parser.parse_args()
if args.frames <= 0:
parser.error("--frames must be positive")
if not args.runtime.is_file():
parser.error(f"--runtime is not a file: {args.runtime}")
if not args.model.is_dir():
parser.error(f"--model is not a directory: {args.model}")
if not args.video.is_file():
parser.error(f"--video is not a file: {args.video}")
if args.output.resolve() == args.video.resolve():
parser.error("--output must not overwrite --video")
return _Arguments(args.runtime, args.model, args.video, args.output, args.frames)
def _preload(video: Path, frame_count: int, input_width: int) -> _PreloadedVideo:
started_at = perf_counter()
capture = cv2.VideoCapture(str(video))
if not capture.isOpened():
raise RuntimeError(f"OpenCV could not open video: {video}")
source_width = int(capture.get(cv2.CAP_PROP_FRAME_WIDTH))
source_height = int(capture.get(cv2.CAP_PROP_FRAME_HEIGHT))
source_fps = float(capture.get(cv2.CAP_PROP_FPS))
if source_width <= 0 or source_height <= 0 or source_fps <= 0.0:
capture.release()
raise RuntimeError(
"video must report positive width, height, and frame rate: "
f"{source_width}x{source_height} at {source_fps} fps"
)
input_height = round(source_height * input_width / source_width)
frames = []
try:
while len(frames) < frame_count:
available, frame = capture.read()
if not available:
break
frames.append(
cv2.resize(frame, (input_width, input_height), interpolation=cv2.INTER_AREA)
)
finally:
capture.release()
if len(frames) != frame_count:
raise RuntimeError(
f"video provided {len(frames)} decodable frames; --frames requested {frame_count}"
)
return _PreloadedVideo(
frames=tuple(frames),
source_width=source_width,
source_height=source_height,
source_fps=source_fps,
decode_resize_ms=(perf_counter() - started_at) * 1000,
)
def _detection_record(result: DetectionResult) -> list[dict[str, object]]:
return [
{
"box": list(detection.box),
"confidence": detection.confidence,
"class_id": detection.class_id,
"class_name": detection.class_name,
}
for detection in result.detections
]
def _run_model(
detector: Yolo26Detector,
video: _PreloadedVideo,
) -> tuple[list[dict[str, object]], dict[str, float | int]]:
detector.detect(video.frames[0], pixel_format="BGR")
records = []
model_seconds = 0.0
preprocess_ms = 0.0
inference_ms = 0.0
postprocess_ms = 0.0
for index, frame in enumerate(video.frames):
started_at = perf_counter()
result = detector.detect(frame, pixel_format="BGR")
model_seconds += perf_counter() - started_at
preprocess_ms += result.timings.preprocess_ms
inference_ms += result.timings.inference_ms
postprocess_ms += result.timings.postprocess_ms
records.append(
{
"frame_index": index,
"timestamp_seconds": index / video.source_fps,
"completed_seconds": model_seconds,
"timings": {
"preprocess_ms": result.timings.preprocess_ms,
"inference_ms": result.timings.inference_ms,
"postprocess_ms": result.timings.postprocess_ms,
},
"detections": _detection_record(result),
}
)
frame_count = len(records)
return records, {
"frames": frame_count,
"model_fps": frame_count / model_seconds,
"preprocess_mean_ms": preprocess_ms / frame_count,
"inference_mean_ms": inference_ms / frame_count,
"postprocess_mean_ms": postprocess_ms / frame_count,
"model_mean_ms": model_seconds * 1000 / frame_count,
"decode_resize_ms": video.decode_resize_ms,
}
def _write_results(
args: _Arguments,
video: _PreloadedVideo,
records: list[dict[str, object]],
summary: dict[str, float | int],
) -> None:
input_height, input_width = video.frames[0].shape[:2]
document = {
"version": 1,
"coordinate_space": "inference_frame_pixels_xyxy",
"source": {
"video": str(args.video.resolve()),
"sha256": _digest(args.video),
"width": video.source_width,
"height": video.source_height,
"fps": video.source_fps,
},
"inference_frame": {"width": input_width, "height": input_height},
"runtime_library": str(args.runtime.resolve()),
"model_directory": str(args.model.resolve()),
"runtime": "yolo26n",
"measurement": {
"decode_and_resize": "preloaded_before_model_timing",
"warmup_frames": 1,
"serialization": "outside_model_timing",
},
"summary": summary,
"frames": records,
}
args.output.parent.mkdir(parents=True, exist_ok=True)
args.output.write_text(json.dumps(document, indent=2, sort_keys=True) + "\n", encoding="utf-8")
def main() -> int:
args = _parse_args()
with Yolo26Detector(runtime=args.runtime, model=args.model) as detector:
video = _preload(args.video, args.frame_count, detector.input_shape[3])
records, summary = _run_model(detector, video)
_write_results(args, video, records, summary)
print(
"Detection performance:\n"
f" throughput: {summary['model_fps']:.2f} FPS\n"
f" detection time: {summary['model_mean_ms']:.2f} ms/frame\n"
f" frames: {summary['frames']}\n"
" mean timing: "
f"preprocess {summary['preprocess_mean_ms']:.2f} ms, "
f"inference {summary['inference_mean_ms']:.2f} ms, "
f"postprocess {summary['postprocess_mean_ms']:.2f} ms\n"
f" video decode and resize: {summary['decode_resize_ms']:.2f} ms total"
)
return 0
if __name__ == "__main__":
raise SystemExit(main())
|