File size: 5,373 Bytes
f17ffc5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import * as ort from "onnxruntime-web";
import type { FleetDetection, FleetHazardType } from "../lib/fleetTelemetry";

const HAZARD_CLASSES: FleetHazardType[] = [
  "pothole",
  "crack",
  "water_logging",
  "construction_zone",
  "traffic_block",
];

type WorkerRequest =
  | { type: "init"; modelUrl: string }
  | { type: "infer"; width: number; height: number; buffer: ArrayBuffer };

type WorkerResponse =
  | { type: "ready" }
  | { type: "result"; detections: FleetDetection[]; inferMs: number }
  | { type: "error"; message: string };

let session: ort.InferenceSession | null = null;

ort.env.wasm.numThreads = 1;

function post(message: WorkerResponse): void {
  self.postMessage(message);
}

async function init(modelUrl: string): Promise<void> {
  if (session) {
    post({ type: "ready" });
    return;
  }
  session = await ort.InferenceSession.create(modelUrl, {
    executionProviders: ["wasm"],
    graphOptimizationLevel: "all",
  });
  post({ type: "ready" });
}

function preprocess(buffer: ArrayBuffer, width: number, height: number): ort.Tensor {
  const pixels = new Uint8ClampedArray(buffer);
  const tensor = new Float32Array(3 * width * height);
  const plane = width * height;

  for (let i = 0; i < plane; i += 1) {
    const src = i * 4;
    tensor[i] = pixels[src] / 255;
    tensor[plane + i] = pixels[src + 1] / 255;
    tensor[plane * 2 + i] = pixels[src + 2] / 255;
  }
  return new ort.Tensor("float32", tensor, [1, 3, height, width]);
}

function clipBox(bbox: [number, number, number, number], size: number): [number, number, number, number] {
  const [x1, y1, x2, y2] = bbox;
  return [
    Math.max(0, Math.min(size, x1)),
    Math.max(0, Math.min(size, y1)),
    Math.max(0, Math.min(size, x2)),
    Math.max(0, Math.min(size, y2)),
  ];
}

function detectionFromRow(row: ArrayLike<number>, offset: number, stride: number, size: number): FleetDetection | null {
  const confidence = Number(row[offset + 4]);
  const classId = Math.round(Number(row[offset + 5]));
  const hazardType = HAZARD_CLASSES[classId];
  if (!hazardType || !Number.isFinite(confidence)) return null;

  const bbox: [number, number, number, number] = clipBox(
    [
      Number(row[offset]),
      Number(row[offset + 1]),
      Number(row[offset + 2]),
      Number(row[offset + 3]),
    ],
    size,
  );
  if (bbox[2] <= bbox[0] || bbox[3] <= bbox[1] || stride < 6) return null;
  return { hazard_type: hazardType, confidence, bbox };
}

function parseNmsOutput(output: ort.Tensor, size: number): FleetDetection[] | null {
  const data = output.data as Float32Array;
  const dims = output.dims;
  let rows = 0;
  let stride = 0;

  if (dims.length === 2 && dims[1] >= 6) {
    rows = dims[0];
    stride = dims[1];
  } else if (dims.length === 3 && dims[2] >= 6) {
    rows = dims[1];
    stride = dims[2];
  }
  if (!rows || !stride) return null;

  const detections: FleetDetection[] = [];
  for (let r = 0; r < rows; r += 1) {
    const det = detectionFromRow(data, r * stride, stride, size);
    if (det && det.confidence > 0.25) detections.push(det);
  }
  return detections;
}

function parseRawYoloOutput(output: ort.Tensor, size: number): FleetDetection[] {
  const data = output.data as Float32Array;
  const dims = output.dims;
  if (dims.length !== 3 || dims[1] < 9) return [];

  const featureCount = dims[1];
  const boxCount = dims[2];
  const detections: FleetDetection[] = [];
  for (let i = 0; i < boxCount; i += 1) {
    let bestClass = -1;
    let bestScore = 0;
    for (let c = 0; c < HAZARD_CLASSES.length; c += 1) {
      const score = data[(4 + c) * boxCount + i];
      if (score > bestScore) {
        bestScore = score;
        bestClass = c;
      }
    }
    const hazardType = HAZARD_CLASSES[bestClass];
    if (!hazardType || bestScore <= 0.25) continue;

    const cx = data[i];
    const cy = data[boxCount + i];
    const w = data[boxCount * 2 + i];
    const h = data[boxCount * 3 + i];
    if (!Number.isFinite(cx + cy + w + h) || featureCount < 9) continue;
    const bbox = clipBox([cx - w / 2, cy - h / 2, cx + w / 2, cy + h / 2], size);
    detections.push({ hazard_type: hazardType, confidence: bestScore, bbox });
  }
  return detections.sort((a, b) => b.confidence - a.confidence).slice(0, 12);
}

function parseDetections(results: Record<string, ort.Tensor>, size: number): FleetDetection[] {
  const output = Object.values(results)[0];
  if (!output) return [];
  return parseNmsOutput(output, size) ?? parseRawYoloOutput(output, size);
}

async function infer(width: number, height: number, buffer: ArrayBuffer): Promise<void> {
  if (!session) throw new Error("Inference worker is not initialized.");
  const started = performance.now();
  const tensor = preprocess(buffer, width, height);
  const inputName = session.inputNames[0];
  const results = (await session.run({ [inputName]: tensor })) as Record<string, ort.Tensor>;
  post({
    type: "result",
    detections: parseDetections(results, width),
    inferMs: Math.round(performance.now() - started),
  });
}

self.addEventListener("message", (event: MessageEvent<WorkerRequest>) => {
  const msg = event.data;
  const task = msg.type === "init" ? init(msg.modelUrl) : infer(msg.width, msg.height, msg.buffer);
  task.catch((err: unknown) => {
    post({ type: "error", message: err instanceof Error ? err.message : "Fleet worker failed." });
  });
});