| |
| |
| |
| |
| |
| |
| |
| |
| |
|
|
| import * as ort from '../vendor/ort/ort.wasm.bundle.min.mjs'; |
|
|
| const IN_W = 320, IN_H = 240; |
| const MODEL_URL = new URL('../vendor/models/version-RFB-320.onnx', import.meta.url); |
|
|
| |
| |
| ort.env.wasm.wasmPaths = new URL('../vendor/ort/', import.meta.url).href; |
| ort.env.wasm.numThreads = 1; |
| ort.env.wasm.proxy = false; |
|
|
| let session = null; |
| let inputName = null; |
|
|
| async function init() { |
| if (session) return; |
| const buf = await (await fetch(MODEL_URL)).arrayBuffer(); |
| session = await ort.InferenceSession.create(new Uint8Array(buf), { |
| executionProviders: ['wasm'], |
| }); |
| inputName = session.inputNames[0]; |
| } |
|
|
| function toTensor(rgba) { |
| const chw = new Float32Array(3 * IN_W * IN_H); |
| const plane = IN_W * IN_H; |
| for (let y = 0; y < IN_H; y++) { |
| for (let x = 0; x < IN_W; x++) { |
| const i = (y * IN_W + x) * 4; |
| const o = y * IN_W + x; |
| chw[o] = (rgba[i] - 127) / 128; |
| chw[plane + o] = (rgba[i + 1] - 127) / 128; |
| chw[2 * plane + o] = (rgba[i + 2] - 127) / 128; |
| } |
| } |
| return new ort.Tensor('float32', chw, [1, 3, IN_H, IN_W]); |
| } |
|
|
| function iou(a, b) { |
| const ix = Math.max(0, Math.min(a.x + a.w, b.x + b.w) - Math.max(a.x, b.x)); |
| const iy = Math.max(0, Math.min(a.y + a.h, b.y + b.h) - Math.max(a.y, b.y)); |
| const inter = ix * iy; |
| const uni = a.w * a.h + b.w * b.h - inter; |
| return uni <= 0 ? 0 : inter / uni; |
| } |
|
|
| function postprocess(scores, boxes, N, origW, origH, thr) { |
| const dets = []; |
| for (let i = 0; i < N; i++) { |
| const p = scores[i * 2 + 1]; |
| if (p < thr) continue; |
| const x1 = boxes[i * 4] * origW, y1 = boxes[i * 4 + 1] * origH; |
| const x2 = boxes[i * 4 + 2] * origW, y2 = boxes[i * 4 + 3] * origH; |
| dets.push({ score: +p.toFixed(3), x: Math.round(x1), y: Math.round(y1), w: Math.round(x2 - x1), h: Math.round(y2 - y1) }); |
| } |
| dets.sort((a, b) => b.score - a.score); |
| const keep = []; |
| for (const d of dets) { |
| if (d.w <= 1 || d.h <= 1) continue; |
| if (!keep.some((k) => iou(k, d) > 0.3)) keep.push(d); |
| } |
| return keep; |
| } |
|
|
| async function detect(rgbaBuf, origW, origH, thr) { |
| await init(); |
| const input = toTensor(new Uint8ClampedArray(rgbaBuf)); |
| const out = await session.run({ [inputName]: input }); |
| const scoresT = out[session.outputNames.find((n) => /score/i.test(n)) || session.outputNames[0]]; |
| const boxesT = out[session.outputNames.find((n) => /box/i.test(n)) || session.outputNames[1]]; |
| const N = scoresT.dims[1]; |
| return postprocess(scoresT.data, boxesT.data, N, origW, origH, thr); |
| } |
|
|
| self.onmessage = async (e) => { |
| const msg = e.data || {}; |
| try { |
| if (msg.type === 'init') { await init(); self.postMessage({ type: 'ready' }); return; } |
| if (msg.type === 'detect') { |
| const faces = await detect(msg.rgba, msg.origW, msg.origH, msg.thr ?? 0.7); |
| self.postMessage({ type: 'result', faces }); |
| } |
| } catch (err) { |
| self.postMessage({ type: 'error', message: String(err && err.message || err) }); |
| } |
| }; |
|
|
