初次提交

src/deploy/facedet/common.ts

@@ -0,0 +1,103 @@
+import { cv } from "../../cv"
+import { ImageSource, Model } from "../common/model"
+
+interface IFaceBoxConstructorOption {
+	x1: number
+	y1: number
+	x2: number
+	y2: number
+	score: number
+	imw: number
+	imh: number
+}
+
+export class FaceBox {
+	#option: IFaceBoxConstructorOption;
+	constructor(option: IFaceBoxConstructorOption) { this.#option = option; }
+
+	public get x1() { return this.#option.x1; }
+	public get y1() { return this.#option.y1; }
+	public get x2() { return this.#option.x2; }
+	public get y2() { return this.#option.y2; }
+	public get centerX() { return this.x1 + this.width / 2; }
+	public get centerY() { return this.y1 + this.height / 2; }
+	public get score() { return this.#option.score; }
+	public get left() { return this.x1; }
+	public get top() { return this.y1; }
+	public get width() { return this.x2 - this.x1; }
+	public get height() { return this.y2 - this.y1; }
+
+	/** 转换成整数 */
+	public toInt() {
+		return new FaceBox({
+			...this.#option,
+			x1: parseInt(this.#option.x1 as any), y1: parseInt(this.#option.y1 as any),
+			x2: parseInt(this.#option.x2 as any), y2: parseInt(this.#option.y2 as any),
+		});
+	}
+
+	/** 转换成正方形 */
+	public toSquare() {
+		const { imw, imh } = this.#option;
+		let size = Math.max(this.width, this.height) / 2;
+		const cx = this.centerX, cy = this.centerY;
+		console.log(this)
+
+		if (cx - size < 0) size = cx;
+		if (cx + size > imw) size = imw - cx;
+		if (cy - size < 0) size = cy;
+		if (cy + size > imh) size = imh - cy;
+
+		return new FaceBox({
+			...this.#option,
+			x1: this.centerX - size, y1: this.centerY - size,
+			x2: this.centerX + size, y2: this.centerY + size,
+		});
+	}
+}
+
+export interface FaceDetectOption {
+	/** 阈值，默认0.5 */
+	threshold?: number
+	/** MNS阈值，默认0.3 */
+	mnsThreshold?: number
+}
+
+export abstract class FaceDetector extends Model {
+
+	protected abstract doPredict(image: cv.Mat, option?: FaceDetectOption): Promise<FaceBox[]>;
+
+	public async predict(image: ImageSource, option?: FaceDetectOption) { return Model.resolveImage(image, im => this.doPredict(im, option)); }
+}
+
+
+export function nms(input: FaceBox[], threshold: number) {
+	if (!input.length) return [];
+	input = input.sort((a, b) => b.score - a.score);
+	const merged = input.map(() => 0);
+	const output: FaceBox[] = [];
+
+	for (let i = 0; i < input.length; i++) {
+		if (merged[i]) continue;
+		output.push(input[i]);
+
+		for (let j = i + 1; j < input.length; j++) {
+			if (merged[j]) continue;
+			const inner_x0 = input[i].x1 > input[j].x1 ? input[i].x1 : input[j].x1;
+			const inner_y0 = input[i].y1 > input[j].y1 ? input[i].y1 : input[j].y1;
+			const inner_x1 = input[i].x2 < input[j].x2 ? input[i].x2 : input[j].x2;
+			const inner_y1 = input[i].y2 < input[j].y2 ? input[i].y2 : input[j].y2;
+			const inner_h = inner_y1 - inner_y0 + 1;
+			const inner_w = inner_x1 - inner_x0 + 1;
+			if (inner_h <= 0 || inner_w <= 0) continue;
+			const inner_area = inner_h * inner_w;
+			const h1 = input[j].y2 - input[j].y1 + 1;
+			const w1 = input[j].x2 - input[j].x1 + 1;
+			const area1 = h1 * w1;
+			const score = inner_area / area1;
+			if (score > threshold) merged[j] = 1;
+		}
+	}
+
+	return output;
+}