Press n or j to go to the next uncovered block, b, p or k for the previous block.
| 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 | /** * TensorFlow.js Boundary Detector * * Lazy-loads the TensorFlow.js model for detecting abacus boundaries * (quadrilateral corners) without requiring ArUco markers. */ import type { QuadCorners, Point } from '@/types/vision' // TensorFlow.js types (dynamically imported) type TFLite = typeof import('@tensorflow/tfjs') type LayersModel = import('@tensorflow/tfjs').LayersModel type GraphModel = import('@tensorflow/tfjs').GraphModel // Model types we support type SupportedModel = LayersModel | GraphModel // Model configuration const MODEL_PATH = '/models/abacus-boundary-detector/model.json' const PREPROCESSING_PATH = '/models/abacus-boundary-detector/preprocessing.json' const INPUT_SIZE = 224 // MobileNetV2 standard input size // Preprocessing configuration (loaded from model directory) interface PreprocessingConfig { edge_enhance?: boolean inpaint_markers?: boolean input_size?: number // New heatmap + DSNT model fields model_type?: 'direct' | 'heatmap_dsnt' heatmap_size?: number // e.g., 56 for 56x56 heatmaps num_corners?: number // always 4 for quadrilateral } // Cached model and TensorFlow instance let tfInstance: TFLite | null = null let modelInstance: SupportedModel | null = null let modelLoadPromise: Promise<SupportedModel | null> | null = null let modelCheckFailed = false let isGraphModel = false let preprocessingConfig: PreprocessingConfig | null = null /** * Lazy load TensorFlow.js */ async function loadTensorFlow(): Promise<TFLite> { if (tfInstance) return tfInstance const tf = await import('@tensorflow/tfjs') await tf.setBackend('webgl') await tf.ready() tfInstance = tf return tf } /** * Load preprocessing configuration from the model directory */ async function loadPreprocessingConfig(): Promise<PreprocessingConfig | null> { if (preprocessingConfig) return preprocessingConfig try { const response = await fetch(PREPROCESSING_PATH) if (!response.ok) { console.warn('[BoundaryDetector] No preprocessing.json found, using defaults') return null } preprocessingConfig = await response.json() console.log('[BoundaryDetector] Loaded preprocessing config:', preprocessingConfig) return preprocessingConfig } catch { console.warn('[BoundaryDetector] Failed to load preprocessing.json, using defaults') return null } } /** * Apply Sobel edge detection using TensorFlow.js operations (runs on GPU) * * This preprocessing helps the model focus on structural edges (like the abacus frame) * rather than colors, textures, and background clutter. */ async function applySobelEdges( tensor: import('@tensorflow/tfjs').Tensor3D ): Promise<import('@tensorflow/tfjs').Tensor3D> { const tf = await loadTensorFlow() // Convert to grayscale using luminance formula: 0.299*R + 0.587*G + 0.114*B const weights = tf.tensor1d([0.299, 0.587, 0.114]) const gray = tf.sum(tf.mul(tensor, weights), -1) // Shape: [H, W] // Sobel kernels (need to be 4D for conv2d: [filterH, filterW, inChannels, outChannels]) const sobelX = tf.tensor4d( [ [-1, 0, 1], [-2, 0, 2], [-1, 0, 1], ].flat(), [3, 3, 1, 1] ) const sobelY = tf.tensor4d( [ [-1, -2, -1], [0, 0, 0], [1, 2, 1], ].flat(), [3, 3, 1, 1] ) // Add batch and channel dimensions for conv2d: [1, H, W, 1] const grayBatched = gray.expandDims(0).expandDims(-1) as import('@tensorflow/tfjs').Tensor4D // Apply Sobel filters const gradX = tf.conv2d(grayBatched, sobelX, 1, 'same') const gradY = tf.conv2d(grayBatched, sobelY, 1, 'same') // Compute gradient magnitude const magnitude = tf.sqrt(tf.add(tf.square(gradX), tf.square(gradY))) // Normalize to [0, 1] range const maxVal = magnitude.max() const normalized = tf.div(magnitude, maxVal.add(1e-7)) // Add epsilon to avoid division by zero // Remove batch dimension and convert to 3-channel: [H, W, 3] const squeezed = normalized.squeeze([0, 3]) as import('@tensorflow/tfjs').Tensor2D const edges3ch = tf.stack( [squeezed, squeezed, squeezed], -1 ) as import('@tensorflow/tfjs').Tensor3D // Clean up intermediate tensors weights.dispose() gray.dispose() sobelX.dispose() sobelY.dispose() grayBatched.dispose() gradX.dispose() gradY.dispose() magnitude.dispose() maxVal.dispose() normalized.dispose() squeezed.dispose() return edges3ch } /** * DSNT: Differentiable Spatial to Numerical Transform * * Converts heatmaps to normalized coordinates using soft-argmax. * This is the inference-time version of the DSNT layer used in training. * * @param heatmaps - Tensor of shape [batch, height, width, num_keypoints] * @param temperature - Softmax temperature (higher = sharper peaks) * @returns Tensor of shape [batch, num_keypoints, 2] with (x, y) coordinates in [0, 1] */ async function dsntDecode( heatmaps: import('@tensorflow/tfjs').Tensor4D, temperature: number = 10 ): Promise<import('@tensorflow/tfjs').Tensor3D> { const tf = await loadTensorFlow() // heatmaps shape: [batch, height, width, num_keypoints] const [batchSize, height, width, numKeypoints] = heatmaps.shape // Create normalized coordinate grids [0, 1] // x_coords: [height, width] where each row is [0, 1/(w-1), 2/(w-1), ..., 1] // y_coords: [height, width] where each column is [0, 1/(h-1), 2/(h-1), ..., 1] const xRange = tf.linspace(0, 1, width) const yRange = tf.linspace(0, 1, height) // Create meshgrid - TF.js meshgrid returns Tensor1D[] but actually produces 2D tensors const meshResult = tf.meshgrid(xRange, yRange) const xGrid = meshResult[0] as unknown as import('@tensorflow/tfjs').Tensor2D const yGrid = meshResult[1] as unknown as import('@tensorflow/tfjs').Tensor2D // Reshape for broadcasting: [1, height, width, 1] const xCoords = xGrid.reshape([1, height, width, 1]) const yCoords = yGrid.reshape([1, height, width, 1]) // Apply softmax to each heatmap channel (spatial softmax) // Reshape to [batch * numKeypoints, height * width] for softmax, then back const heatmapsReshaped = heatmaps.transpose([0, 3, 1, 2]) // [batch, keypoints, h, w] const heatmapsFlat = heatmapsReshaped.reshape([batchSize * numKeypoints, height * width]) // Scale by temperature before softmax (higher temp = sharper distribution) const heatmapsScaled = heatmapsFlat.mul(temperature) const heatmapsSoftmax = tf.softmax(heatmapsScaled, 1) // Reshape back to [batch, keypoints, height, width], then [batch, height, width, keypoints] const heatmapsNorm = heatmapsSoftmax .reshape([batchSize, numKeypoints, height, width]) .transpose([0, 2, 3, 1]) // Back to [batch, h, w, keypoints] // Compute expected x and y values (weighted sum of coordinates) // result: [batch, 1, 1, keypoints] for each coordinate const xExpected = tf.sum(tf.mul(heatmapsNorm, xCoords), [1, 2]) // [batch, keypoints] const yExpected = tf.sum(tf.mul(heatmapsNorm, yCoords), [1, 2]) // [batch, keypoints] // Stack to get [batch, keypoints, 2] with (x, y) pairs const coordinates = tf.stack([xExpected, yExpected], 2) as import('@tensorflow/tfjs').Tensor3D // Clean up intermediate tensors xRange.dispose() yRange.dispose() xGrid.dispose() yGrid.dispose() xCoords.dispose() yCoords.dispose() heatmapsReshaped.dispose() heatmapsFlat.dispose() heatmapsScaled.dispose() heatmapsSoftmax.dispose() heatmapsNorm.dispose() xExpected.dispose() yExpected.dispose() return coordinates } /** * Check if the model file exists before attempting to load it */ async function checkModelExists(): Promise<boolean> { if (modelCheckFailed) return false try { const response = await fetch(MODEL_PATH, { method: 'HEAD' }) if (!response.ok) { modelCheckFailed = true return false } return true } catch { modelCheckFailed = true return false } } /** * Lazy load the boundary detection model * Returns null if model doesn't exist (not yet trained) */ async function loadModel(): Promise<SupportedModel | null> { console.log('[BoundaryDetector] loadModel called') if (modelInstance) { console.log('[BoundaryDetector] Returning cached model instance') return modelInstance } if (modelCheckFailed) { console.log('[BoundaryDetector] Model check previously failed, returning null') return null } if (modelLoadPromise) { console.log('[BoundaryDetector] Returning existing load promise') return modelLoadPromise } modelLoadPromise = (async () => { console.log('[BoundaryDetector] Checking if model exists...') const exists = await checkModelExists() if (!exists) { console.warn( '[BoundaryDetector] Model not found at', MODEL_PATH, '- Marker-free calibration disabled. Train the boundary detector first.' ) return null } console.log('[BoundaryDetector] Model file exists, loading preprocessing config...') // Load preprocessing config (non-blocking, defaults to no edge enhancement) await loadPreprocessingConfig() console.log('[BoundaryDetector] Loading TensorFlow.js...') const tf = await loadTensorFlow() console.log('[BoundaryDetector] TensorFlow.js loaded, attempting to load model...') const startTime = performance.now() let model: SupportedModel | null = null // Try loading as graph model first (preferred format) try { console.log('[BoundaryDetector] Trying loadGraphModel...') model = await tf.loadGraphModel(MODEL_PATH) isGraphModel = true console.log('[BoundaryDetector] Successfully loaded as GraphModel') } catch (graphErr) { console.warn('[BoundaryDetector] loadGraphModel failed:', graphErr) // Fall back to layers model try { console.log('[BoundaryDetector] Trying loadLayersModel...') model = await tf.loadLayersModel(MODEL_PATH) isGraphModel = false console.log('[BoundaryDetector] Successfully loaded as LayersModel') } catch (layersErr) { console.error('[BoundaryDetector] loadLayersModel also failed:', layersErr) console.error('[BoundaryDetector] Failed to load model with either method') modelLoadPromise = null modelCheckFailed = true return null } } const loadTime = performance.now() - startTime console.log( `[BoundaryDetector] Model loaded in ${loadTime.toFixed(0)}ms, isGraphModel=${isGraphModel}` ) // Log model info if (model) { if (isGraphModel) { const graphModel = model as GraphModel console.log('[BoundaryDetector] GraphModel inputs:', graphModel.inputs) console.log('[BoundaryDetector] GraphModel outputs:', graphModel.outputs) } else { const layersModel = model as LayersModel console.log('[BoundaryDetector] LayersModel inputs:', layersModel.inputs) console.log('[BoundaryDetector] LayersModel outputs:', layersModel.outputs) } } modelInstance = model return model })() return modelLoadPromise } /** * Preprocess an image for boundary detection * * @param imageData - Raw image data from canvas * @returns Preprocessed tensor ready for inference */ async function preprocessImage(imageData: ImageData): Promise<import('@tensorflow/tfjs').Tensor4D> { const tf = await loadTensorFlow() // Convert ImageData to tensor (RGB) const tensor = tf.browser.fromPixels(imageData, 3) // Resize to model input size const resized = tf.image.resizeBilinear(tensor, [ INPUT_SIZE, INPUT_SIZE, ]) as import('@tensorflow/tfjs').Tensor3D // Normalize to [0,1] first let normalized = resized.div(255) as import('@tensorflow/tfjs').Tensor3D // Apply Sobel edge detection if the model was trained with it if (preprocessingConfig?.edge_enhance) { console.log('[BoundaryDetector] Applying Sobel edge enhancement') const edges = await applySobelEdges(normalized) normalized.dispose() normalized = edges } // Add batch dimension const batched = normalized.expandDims(0) as import('@tensorflow/tfjs').Tensor4D // Clean up intermediate tensors tensor.dispose() resized.dispose() normalized.dispose() return batched } /** * Result of boundary detection */ export interface BoundaryDetectionResult { /** Detected corners in normalized coordinates (0-1) */ corners: QuadCorners /** Confidence score (0-1) based on prediction consistency */ confidence: number } /** * Detect abacus boundary corners from an image * * @param imageData - RGB image data from video frame * @returns Detection result with corners, or null if model not available */ export async function detectBoundary( imageData: ImageData ): Promise<BoundaryDetectionResult | null> { console.log( '[BoundaryDetector] detectBoundary called, imageData size:', imageData.width, 'x', imageData.height ) const model = await loadModel() if (!model) { console.log('[BoundaryDetector] Model not available, returning null') return null } // Preprocess console.log('[BoundaryDetector] Preprocessing image...') const input = await preprocessImage(imageData) console.log('[BoundaryDetector] Input tensor shape:', input.shape) // Debug: verify input normalization (use tf.min/max to avoid stack overflow on large arrays) const tf = await loadTensorFlow() const inputMin = (await tf.min(input).data())[0] const inputMax = (await tf.max(input).data())[0] const inputMean = (await tf.mean(input).data())[0] console.log( `[BoundaryDetector] Input stats: min=${inputMin.toFixed(4)}, max=${inputMax.toFixed(4)}, mean=${inputMean.toFixed(4)} (expected: 0-1 range)` ) // Run inference console.log('[BoundaryDetector] Running inference, isGraphModel:', isGraphModel) let output: import('@tensorflow/tfjs').Tensor if (isGraphModel) { const graphModel = model as GraphModel console.log('[BoundaryDetector] Using GraphModel.predict()') output = graphModel.predict(input) as import('@tensorflow/tfjs').Tensor } else { const layersModel = model as LayersModel console.log('[BoundaryDetector] Using LayersModel.predict()') output = layersModel.predict(input) as import('@tensorflow/tfjs').Tensor } console.log('[BoundaryDetector] Output tensor shape:', output.shape) let corners: QuadCorners // Determine model type from preprocessing config or output shape const modelType = preprocessingConfig?.model_type ?? 'direct' const isHeatmapModel = modelType === 'heatmap_dsnt' || output.shape.length === 4 if (isHeatmapModel) { // Heatmap + DSNT model: output is [batch, height, width, 4] console.log('[BoundaryDetector] Using heatmap + DSNT decoding') // Apply DSNT to convert heatmaps to coordinates const heatmaps = output as import('@tensorflow/tfjs').Tensor4D // Debug: log heatmap statistics to diagnose detection issues (use tf.min/max to avoid stack overflow) const minVal = (await tf.min(heatmaps).data())[0] const maxVal = (await tf.max(heatmaps).data())[0] const meanVal = (await tf.mean(heatmaps).data())[0] console.log( `[BoundaryDetector] Heatmap stats: min=${minVal.toFixed(4)}, max=${maxVal.toFixed(4)}, mean=${meanVal.toFixed(4)}` ) // Log per-corner max values and diagnose (use tf.min/max to avoid stack overflow) const [_, h, w, numCorners] = heatmaps.shape const cornerMaxes: number[] = [] for (let c = 0; c < numCorners; c++) { const cornerHeatmap = tf.slice(heatmaps, [0, 0, 0, c], [1, h, w, 1]) const cornerMax = (await tf.max(cornerHeatmap).data())[0] const cornerMean = (await tf.mean(cornerHeatmap).data())[0] cornerMaxes.push(cornerMax) console.log( `[BoundaryDetector] Corner ${c} heatmap: max=${cornerMax.toFixed(4)}, mean=${cornerMean.toFixed(4)}, ratio=${(cornerMax / cornerMean).toFixed(1)}x` ) cornerHeatmap.dispose() } // Diagnostic summary const avgCornerMax = cornerMaxes.reduce((a, b) => a + b, 0) / cornerMaxes.length if (avgCornerMax < 0.1) { console.warn( `[BoundaryDetector] ⚠️ DIAGNOSIS: Heatmap peaks are very weak (avg max=${avgCornerMax.toFixed(4)}). Model may not be detecting corners in this image.` ) } else if (maxVal - minVal < 0.1) { console.warn( `[BoundaryDetector] ⚠️ DIAGNOSIS: Heatmaps are nearly flat (range=${(maxVal - minVal).toFixed(4)}). No clear corner detections.` ) } else { console.log( `[BoundaryDetector] ✓ Heatmaps look reasonable (avg max=${avgCornerMax.toFixed(4)}, range=${(maxVal - minVal).toFixed(4)})` ) } const coords = await dsntDecode(heatmaps) // coords shape: [batch, 4, 2] where each keypoint is (x, y) const coordsData = await coords.data() console.log('[BoundaryDetector] DSNT decoded coordinates:', Array.from(coordsData)) // Parse corners - order: topLeft, topRight, bottomLeft, bottomRight corners = { topLeft: { x: coordsData[0], y: coordsData[1] }, topRight: { x: coordsData[2], y: coordsData[3] }, bottomLeft: { x: coordsData[4], y: coordsData[5] }, bottomRight: { x: coordsData[6], y: coordsData[7] }, } coords.dispose() } else { // Direct regression model: output is [batch, 8] console.log('[BoundaryDetector] Using direct coordinate output') const predictions = await output.data() console.log('[BoundaryDetector] Raw predictions:', Array.from(predictions)) // Parse corner coordinates // Training data order: topLeft, topRight, bottomLeft, bottomRight corners = { topLeft: { x: predictions[0], y: predictions[1] }, topRight: { x: predictions[2], y: predictions[3] }, bottomLeft: { x: predictions[4], y: predictions[5] }, bottomRight: { x: predictions[6], y: predictions[7] }, } } console.log('[BoundaryDetector] Parsed corners:', corners) // Compute confidence based on geometric validity const confidence = computeConfidence(corners) console.log('[BoundaryDetector] Confidence:', confidence) // Clean up tensors input.dispose() output.dispose() return { corners, confidence } } /** * Compute a confidence score based on geometric validity of detected corners * * Valid abacus boundaries should form a convex quadrilateral. * Note: Due to perspective distortion, corners may not form a perfect rectangle. * The top may be wider/narrower than the bottom, etc. */ function computeConfidence(corners: QuadCorners): number { let confidence = 1.0 // Check that corners are in valid range (0-1) const allCorners = [corners.topLeft, corners.topRight, corners.bottomRight, corners.bottomLeft] for (const corner of allCorners) { if (corner.x < 0 || corner.x > 1 || corner.y < 0 || corner.y > 1) { confidence *= 0.5 } } // Check basic ordering within each edge (allows for perspective): // - topLeft.x should be less than topRight.x (top edge goes left to right) // - bottomLeft.x should be less than bottomRight.x (bottom edge goes left to right) // - topLeft.y should be less than bottomLeft.y OR topRight.y less than bottomRight.y // (at least one side should have top above bottom - allows for tilted cameras) if (corners.topLeft.x >= corners.topRight.x) confidence *= 0.7 if (corners.bottomLeft.x >= corners.bottomRight.x) confidence *= 0.7 // Check that the quadrilateral has some vertical extent // (either left side or right side should have top above bottom) const leftVertical = corners.bottomLeft.y - corners.topLeft.y const rightVertical = corners.bottomRight.y - corners.topRight.y if (leftVertical <= 0 && rightVertical <= 0) { // Both sides are inverted - definitely wrong confidence *= 0.5 } // Check for minimum size (quadrilateral shouldn't be tiny) const topWidth = Math.abs(corners.topRight.x - corners.topLeft.x) const bottomWidth = Math.abs(corners.bottomRight.x - corners.bottomLeft.x) const avgWidth = (topWidth + bottomWidth) / 2 const avgHeight = (Math.abs(leftVertical) + Math.abs(rightVertical)) / 2 if (avgWidth < 0.05 || avgHeight < 0.05) { // Too small - likely a failed detection confidence *= 0.3 } // Check that quadrilateral is convex using cross product // A convex quadrilateral has all cross products with the same sign const isConvex = checkConvexity(corners) if (!isConvex) { confidence *= 0.6 } return Math.max(0, Math.min(1, confidence)) } /** * Check if a quadrilateral is convex using cross products */ function checkConvexity(corners: QuadCorners): boolean { const points = [corners.topLeft, corners.topRight, corners.bottomRight, corners.bottomLeft] let sign = 0 for (let i = 0; i < 4; i++) { const p1 = points[i] const p2 = points[(i + 1) % 4] const p3 = points[(i + 2) % 4] // Cross product of vectors (p1->p2) and (p2->p3) const cross = (p2.x - p1.x) * (p3.y - p2.y) - (p2.y - p1.y) * (p3.x - p2.x) if (i === 0) { sign = Math.sign(cross) } else if (Math.sign(cross) !== sign && cross !== 0) { return false // Not convex } } return true } /** * Denormalize corners from [0,1] to pixel coordinates */ export function denormalizeCorners( corners: QuadCorners, width: number, height: number ): QuadCorners { const denormalize = (p: Point): Point => ({ x: p.x * width, y: p.y * height, }) return { topLeft: denormalize(corners.topLeft), topRight: denormalize(corners.topRight), bottomRight: denormalize(corners.bottomRight), bottomLeft: denormalize(corners.bottomLeft), } } /** * Normalize corners from pixel coordinates to [0,1] */ export function normalizeCorners(corners: QuadCorners, width: number, height: number): QuadCorners { const normalize = (p: Point): Point => ({ x: p.x / width, y: p.y / height, }) return { topLeft: normalize(corners.topLeft), topRight: normalize(corners.topRight), bottomRight: normalize(corners.bottomRight), bottomLeft: normalize(corners.bottomLeft), } } /** * Check if the model is currently loaded */ export function isModelLoaded(): boolean { return modelInstance !== null } /** * Check if model loading has failed (model doesn't exist) */ export function isModelUnavailable(): boolean { return modelCheckFailed } /** * Preload the model (for eager initialization) * Returns true if model loaded successfully, false if unavailable */ export async function preloadModel(): Promise<boolean> { const model = await loadModel() return model !== null } /** * Dispose of the model to free memory */ export function disposeModel(): void { if (modelInstance) { modelInstance.dispose() modelInstance = null modelLoadPromise = null } } /** * Reset the model state to allow retrying after a failure * Useful when the model file has been updated */ export function resetModelState(): void { if (modelInstance) { modelInstance.dispose() } modelInstance = null modelLoadPromise = null modelCheckFailed = false preprocessingConfig = null } /** * Get model input dimensions */ export function getModelInputSize(): { width: number; height: number } { return { width: INPUT_SIZE, height: INPUT_SIZE } } /** * Check if edge enhancement preprocessing is enabled for the loaded model */ export function isEdgeEnhanceEnabled(): boolean { return preprocessingConfig?.edge_enhance ?? false } /** * Get the current preprocessing configuration */ export function getPreprocessingConfig(): PreprocessingConfig | null { return preprocessingConfig } /** * Get the model type (direct regression vs heatmap+DSNT) */ export function getModelType(): 'direct' | 'heatmap_dsnt' | null { return preprocessingConfig?.model_type ?? null } /** * Check if the model uses heatmap + DSNT architecture */ export function isHeatmapModel(): boolean { return preprocessingConfig?.model_type === 'heatmap_dsnt' } |