Repository navigation
Expand file tree
/
Copy pathmain.ts
More file actions
419 lines (380 loc) · 13.2 KB
/
Copy pathmain.ts
File metadata and controls
419 lines (380 loc) · 13.2 KB
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
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
import { sampleGallery } from "./generated/sample-gallery";
import { getDefaultModelProfile, getModelProfile, modelProfiles } from "./modelProfiles";
import { pixelsToCHWTensorData } from "./preprocess";
import "./styles.css";
type ApiSearchResult = {
id: string;
label: string;
floor?: string;
zone?: string;
x?: number;
y?: number;
imagePath?: string;
score: number;
rank: number;
};
type ApiSearchResponse = {
modelId: string;
topK: number;
confidence?: {
level: "confident" | "ambiguous" | "uncertain";
topScore: number;
margin: number;
reason: string;
};
results: ApiSearchResult[];
};
type InferenceTimings = {
preprocessMs: number;
modelLoadMs: number;
sessionCreateMs: number;
inferenceMs: number;
apiMs: number;
totalMs: number;
};
type EmbeddingResult = {
embedding: number[];
backend: string;
cacheHit: boolean;
sessionReused: boolean;
fallbackErrors: Array<{ backend: string; error: string }>;
timings: Omit<InferenceTimings, "apiMs" | "totalMs">;
};
let modelProfile = getDefaultModelProfile();
const SAMPLE_QUERY = "/samples/query-starfield-north.jpg";
const app = document.querySelector<HTMLDivElement>("#app");
if (!app) {
throw new Error("앱 루트 요소를 찾을 수 없습니다.");
}
app.innerHTML = `
<section class="shell">
<header class="header">
<div>
<p class="eyebrow">COEX visual localization prototype</p>
<h1>path-finder</h1>
</div>
<span id="modelBadge" class="badge">ONNX CPU/WASM · ${modelProfile.dimensions}D</span>
</header>
<section class="workspace">
<div class="panel query-panel">
<div class="panel-title">
<h2>입력 이미지</h2>
<button id="sampleButton" type="button">샘플 실행</button>
</div>
<label class="model-select">
<span>모델</span>
<select id="modelSelect">
${modelProfiles
.map(
(profile) =>
`<option value="${profile.id}">${profile.label} ${profile.dimensions}D</option>`,
)
.join("")}
</select>
</label>
<img id="queryImage" alt="검색할 샘플 이미지" src="${SAMPLE_QUERY}" />
<label class="file-input">
<span>이미지 선택</span>
<input id="fileInput" type="file" accept="image/*" />
</label>
<p id="status" class="status">모델을 준비하는 중입니다.</p>
</div>
<div class="panel results-panel">
<div class="panel-title">
<h2>위치 후보</h2>
<span id="latency">-</span>
</div>
<p id="confidence" class="confidence">-</p>
<dl id="metrics" class="metrics"></dl>
<ol id="results" class="results"></ol>
</div>
</section>
<section class="gallery-strip" aria-label="샘플 지도 이미지">
${sampleGallery
.map(
(item) => `
<article>
<img src="${item.imagePath}" alt="${item.label}" />
<strong>${item.label}</strong>
<span>${item.floor} · ${item.zone}</span>
</article>
`,
)
.join("")}
</section>
</section>
`;
const queryImage = document.querySelector<HTMLImageElement>("#queryImage")!;
const fileInput = document.querySelector<HTMLInputElement>("#fileInput")!;
const modelSelect = document.querySelector<HTMLSelectElement>("#modelSelect")!;
const sampleButton = document.querySelector<HTMLButtonElement>("#sampleButton")!;
const statusEl = document.querySelector<HTMLParagraphElement>("#status")!;
const latencyEl = document.querySelector<HTMLSpanElement>("#latency")!;
const resultsEl = document.querySelector<HTMLOListElement>("#results")!;
const modelBadge = document.querySelector<HTMLSpanElement>("#modelBadge")!;
const metricsEl = document.querySelector<HTMLDListElement>("#metrics")!;
const confidenceEl = document.querySelector<HTMLParagraphElement>("#confidence")!;
let inferenceWorker: Worker | null = null;
let inferenceWorkerKind: "wasm" | "webgpu" | null = null;
function getInferenceWorker(): Worker {
const nextKind = modelProfile.backendCandidates.includes("webgpu") ? "webgpu" : "wasm";
if (!inferenceWorker || inferenceWorkerKind !== nextKind) {
inferenceWorker?.terminate();
inferenceWorker =
nextKind === "webgpu"
? new Worker(new URL("./webgpuInferenceWorker.ts", import.meta.url), { type: "module" })
: new Worker(new URL("./inferenceWorker.ts", import.meta.url), { type: "module" });
inferenceWorkerKind = nextKind;
}
return inferenceWorker;
}
async function imageToTensorData(image: HTMLImageElement): Promise<Float32Array> {
if (!image.complete || image.naturalWidth === 0) {
await image.decode();
}
const canvas = document.createElement("canvas");
canvas.width = modelProfile.inputSize;
canvas.height = modelProfile.inputSize;
const context = canvas.getContext("2d", { willReadFrequently: true });
if (!context) {
throw new Error("캔버스 컨텍스트를 만들 수 없습니다.");
}
context.drawImage(image, 0, 0, modelProfile.inputSize, modelProfile.inputSize);
const pixels = context.getImageData(0, 0, modelProfile.inputSize, modelProfile.inputSize).data;
return pixelsToCHWTensorData(pixels, modelProfile);
}
function normalizeEmbedding(values: Iterable<number>): number[] {
const vector = Array.from(values);
const norm = Math.hypot(...vector);
if (norm === 0) {
return vector;
}
return vector.map((value) => value / norm);
}
async function embedCurrentImage(): Promise<EmbeddingResult> {
const preprocessStartedAt = performance.now();
const input = await imageToTensorData(queryImage);
const preprocessMs = performance.now() - preprocessStartedAt;
const worker = getInferenceWorker();
return new Promise((resolve, reject) => {
const id = crypto.randomUUID();
const timeout = window.setTimeout(() => {
worker.terminate();
if (inferenceWorker === worker) {
inferenceWorker = null;
inferenceWorkerKind = null;
}
reject(
new Error(
`${modelProfile.label} 브라우저 추론이 ${Math.round(
modelProfile.inferenceTimeoutMs / 1000,
)}초를 초과했습니다.`,
),
);
}, modelProfile.inferenceTimeoutMs);
const MessageEvent) => {
const message = event.data as
| {
type: "embedding";
id: string;
values: number[];
backend: string;
cacheHit: boolean;
sessionReused: boolean;
fallbackErrors: Array<{ backend: string; error: string }>;
timings: {
modelLoadMs: number;
sessionCreateMs: number;
inferenceMs: number;
};
}
| { type: "error"; id: string; error: string };
if (message.id !== id) {
return;
}
window.clearTimeout(timeout);
worker.removeEventListener("message", onMessage);
if (message.type === "error") {
reject(new Error(message.error));
return;
}
const result = {
embedding: normalizeEmbedding(message.values),
backend: message.backend,
cacheHit: message.cacheHit,
sessionReused: message.sessionReused,
fallbackErrors: message.fallbackErrors,
timings: {
preprocessMs,
modelLoadMs: message.timings.modelLoadMs,
sessionCreateMs: message.timings.sessionCreateMs,
inferenceMs: message.timings.inferenceMs,
},
};
if (message.backend === "webgpu") {
worker.terminate();
if (inferenceWorker === worker) {
inferenceWorker = null;
inferenceWorkerKind = null;
}
}
resolve(result);
};
worker.addEventListener("message", onMessage);
worker.postMessage(
{
type: "embed",
id,
modelUrl: modelProfile.modelUrl,
inputSize: modelProfile.inputSize,
backendCandidates: modelProfile.backendCandidates,
input,
},
[input.buffer],
);
});
}
async function search(embedding: number[]): Promise<ApiSearchResponse> {
const response = await fetch("/api/search", {
method: "POST",
headers: { "content-type": "application/json" },
body: JSON.stringify({ embedding, topK: 5, modelId: modelProfile.searchModelId ?? modelProfile.id }),
});
if (!response.ok) {
throw new Error(`검색 API 오류: ${response.status}`);
}
return (await response.json()) as ApiSearchResponse;
}
function renderResults(data: ApiSearchResponse): void {
const confidence = data.confidence;
confidenceEl.textContent = confidence
? `${confidence.level} · top ${(confidence.topScore * 100).toFixed(1)}% · margin ${(
confidence.margin * 100
).toFixed(1)}%`
: "-";
confidenceEl.dataset.level = confidence?.level ?? "";
resultsEl.innerHTML = data.results
.map(
(result) => `
<li>
<img src="${result.imagePath}" alt="${result.label}" />
<div>
<strong>${result.rank}. ${result.label}</strong>
<span>${result.floor ?? "-"} · ${result.zone ?? "-"} · (${result.x ?? 0}, ${
result.y ?? 0
})</span>
</div>
<b>${(result.score * 100).toFixed(1)}%</b>
</li>
`,
)
.join("");
}
function formatMs(value: number): string {
return `${value.toFixed(1)}ms`;
}
function renderMetrics(data: EmbeddingResult, apiMs: number, totalMs: number): void {
const timings: InferenceTimings = {
...data.timings,
apiMs,
totalMs,
};
const fallbackText =
data.fallbackErrors.length > 0
? data.fallbackErrors.map((item) => `${item.backend}: ${item.error}`).join(" / ")
: "없음";
const cacheText = data.sessionReused ? "session reused" : data.cacheHit ? "hit" : "miss";
metricsEl.innerHTML = `
<div><dt>백엔드</dt><dd>${data.backend}</dd></div>
<div><dt>모델 캐시</dt><dd>${cacheText}</dd></div>
<div><dt>전처리</dt><dd>${formatMs(timings.preprocessMs)}</dd></div>
<div><dt>모델 로드</dt><dd>${formatMs(timings.modelLoadMs)}</dd></div>
<div><dt>세션 생성</dt><dd>${formatMs(timings.sessionCreateMs)}</dd></div>
<div><dt>추론</dt><dd>${formatMs(timings.inferenceMs)}</dd></div>
<div><dt>API</dt><dd>${formatMs(timings.apiMs)}</dd></div>
<div><dt>fallback</dt><dd>${fallbackText}</dd></div>
`;
}
function setModelAvailability(): void {
modelBadge.textContent = `${modelProfile.backendCandidates.join("→")} · ${
modelProfile.dimensions
}D`;
sampleButton.disabled = !modelProfile.browserRunnable;
fileInput.disabled = !modelProfile.browserRunnable;
}
function showDisabledModelState(): void {
resultsEl.innerHTML = "";
confidenceEl.textContent = "-";
confidenceEl.dataset.level = "";
metricsEl.innerHTML = "";
latencyEl.textContent = "-";
statusEl.textContent =
modelProfile.disabledReason ?? "이 모델은 브라우저 실행을 지원하지 않습니다.";
}
async function runSearch(): Promise<void> {
if (!modelProfile.browserRunnable) {
showDisabledModelState();
return;
}
statusEl.textContent = "브라우저에서 ONNX 임베딩을 계산하는 중입니다.";
resultsEl.innerHTML = "";
confidenceEl.textContent = "-";
confidenceEl.dataset.level = "";
metricsEl.innerHTML = "";
const start = performance.now();
try {
const embeddingResult = await embedCurrentImage();
const apiStartedAt = performance.now();
const data = await search(embeddingResult.embedding);
const apiMs = performance.now() - apiStartedAt;
const elapsed = performance.now() - start;
latencyEl.textContent = `${elapsed.toFixed(1)}ms`;
statusEl.textContent = "검색이 완료되었습니다.";
renderMetrics(embeddingResult, apiMs, elapsed);
renderResults(data);
} catch (error) {
statusEl.textContent = error instanceof Error ? error.message : "검색 중 오류가 발생했습니다.";
latencyEl.textContent = "-";
}
}
fileInput.addEventListener("change", () => {
const file = fileInput.files?.[0];
if (!file) {
return;
}
queryImage.src = URL.createObjectURL(file);
queryImage. => void runSearch();
});
modelSelect.addEventListener("change", () => {
const nextProfile = getModelProfile(modelSelect.value);
if (!nextProfile) {
return;
}
modelProfile = nextProfile;
inferenceWorker?.terminate();
inferenceWorker = null;
inferenceWorkerKind = null;
setModelAvailability();
if (!modelProfile.browserRunnable) {
showDisabledModelState();
return;
}
resultsEl.innerHTML = "";
latencyEl.textContent = "-";
statusEl.textContent = `${modelProfile.label} 모델을 준비하는 중입니다.`;
if (modelProfile.autoRunOnSelect === false) {
statusEl.textContent = `${modelProfile.label}은 실험 옵션입니다. 샘플 실행 버튼으로 직접 실행하세요.`;
return;
}
void runSearch();
});
sampleButton.addEventListener("click", () => {
queryImage.src = SAMPLE_QUERY;
queryImage. => void runSearch();
if (queryImage.complete) {
void runSearch();
}
});
statusEl.textContent = "모델 준비가 끝났습니다. 샘플을 실행합니다.";
setModelAvailability();
void runSearch();