d2fbc413b3
* feat(canvas): 短剧项目空间 - 建立项目章节与画布关联 MVP * feat(canvas): 短剧项目空间 - 完善项目资产、镜头与流程闭环 Refs #30 * fix(deploy): 本地部署 - 修复前端构建并忽略本地 PostgreSQL 配置 * feat(*): 短剧项目空间 - 完成项目、章节、资产、角色与画布创作闭环 Refs #30 --------- Co-authored-by: ddCat <ddcat666@88.com>
86 lines
2.8 KiB
TypeScript
86 lines
2.8 KiB
TypeScript
/// <reference lib="webworker" />
|
||
|
||
import { FaceDetector } from "@mediapipe/tasks-vision";
|
||
|
||
import type { CanvasFaceBox } from "./canvas-emotion";
|
||
|
||
type DetectFaceRequest = {
|
||
id: number;
|
||
image: ImageBitmap;
|
||
};
|
||
|
||
type DetectFaceResponse = {
|
||
id: number;
|
||
faces?: CanvasFaceBox[];
|
||
imageWidth?: number;
|
||
imageHeight?: number;
|
||
error?: string;
|
||
};
|
||
|
||
let detectorPromise: Promise<FaceDetector> | null = null;
|
||
|
||
const workerGlobal = self as typeof self & {
|
||
importScripts: (...urls: string[]) => void;
|
||
import?: (url: string) => Promise<unknown>;
|
||
};
|
||
|
||
// MediaPipe 会在模块 Worker 中从 importScripts 失败分支转向 self.import。
|
||
// 这里显式标记为运行时 URL，避免 Vite 将 public loader 当作源码模块转换。
|
||
workerGlobal.importScripts = () => { throw new TypeError("module worker uses dynamic import"); };
|
||
workerGlobal.import = async (url: string) => {
|
||
const response = await fetch(url.replace(/\?import(?:&.*)?$/, ""));
|
||
if (!response.ok) throw new Error(`MediaPipe loader 加载失败：${response.status}`);
|
||
const blobUrl = URL.createObjectURL(new Blob([await response.text()], { type: "text/javascript" }));
|
||
try {
|
||
return await import(/* @vite-ignore */ blobUrl);
|
||
} finally {
|
||
URL.revokeObjectURL(blobUrl);
|
||
}
|
||
};
|
||
|
||
function getDetector() {
|
||
if (!detectorPromise) {
|
||
detectorPromise = FaceDetector.createFromOptions(
|
||
{
|
||
wasmLoaderPath: "/mediapipe/wasm/vision_wasm_module_internal.js",
|
||
wasmBinaryPath: "/mediapipe/wasm/vision_wasm_module_internal.wasm",
|
||
},
|
||
{
|
||
baseOptions: { modelAssetPath: "/canvas/models/blaze-face-full-range-sparse.tflite" },
|
||
runningMode: "IMAGE",
|
||
minDetectionConfidence: 0.25,
|
||
minSuppressionThreshold: 0.3,
|
||
},
|
||
);
|
||
}
|
||
return detectorPromise;
|
||
}
|
||
|
||
self.onmessage = async (event: MessageEvent<DetectFaceRequest>) => {
|
||
const { id, image } = event.data;
|
||
const response: DetectFaceResponse = { id, imageWidth: image.width, imageHeight: image.height };
|
||
try {
|
||
const detector = await getDetector();
|
||
response.faces = detector.detect(image).detections.flatMap((detection, index) => {
|
||
const box = detection.boundingBox;
|
||
if (!box) return [];
|
||
return [{
|
||
id: `face-${id}-${index}`,
|
||
x: box.originX,
|
||
y: box.originY,
|
||
width: box.width,
|
||
height: box.height,
|
||
confidence: detection.categories[0]?.score,
|
||
source: "detected" as const,
|
||
}];
|
||
});
|
||
} catch (error) {
|
||
response.error = error instanceof Error ? error.message : "人脸识别失败";
|
||
} finally {
|
||
image.close();
|
||
}
|
||
self.postMessage(response);
|
||
};
|
||
|
||
export {};
|