Spaces:
Sleeping
Sleeping
File size: 34,294 Bytes
9f3acc0 392e60f 433e26f 0332541 433e26f 0332541 433e26f 392e60f 0f5dd41 4ed47f2 392e60f 433e26f f069dfc 9f3acc0 f069dfc 9f3acc0 0332541 0f5dd41 5c6d1cb 0f5dd41 5c6d1cb 0f5dd41 5c6d1cb 0f5dd41 5c6d1cb 0f5dd41 5c6d1cb 0f5dd41 5c6d1cb 0f5dd41 5c6d1cb 0f5dd41 5c6d1cb 0f5dd41 5c6d1cb 0f5dd41 5c6d1cb 0f5dd41 0332541 5c6d1cb 392e60f 4ed47f2 392e60f b55ddea 9f3acc0 b55ddea 9f3acc0 b55ddea 9f3acc0 433e26f 9f3acc0 0332541 392e60f 0f5dd41 392e60f 9f3acc0 433e26f 0332541 433e26f 0f5dd41 433e26f 9f3acc0 433e26f 0f5dd41 392e60f 0f5dd41 4ed47f2 5c6d1cb 9f3acc0 4ed47f2 392e60f 433e26f 392e60f 433e26f 9f3acc0 392e60f 433e26f 392e60f 5c6d1cb 0332541 0f5dd41 1654955 0f5dd41 aa29cac 1654955 aa29cac 1654955 aa29cac 1654955 aa29cac 1654955 aa29cac 0f5dd41 fedb187 392e60f 433e26f 9f3acc0 392e60f 9f3acc0 392e60f 9f3acc0 392e60f 433e26f 0f5dd41 433e26f 9f3acc0 392e60f 9f3acc0 392e60f 433e26f 0f5dd41 433e26f 9f3acc0 433e26f 9f3acc0 392e60f 9f3acc0 0332541 9f3acc0 5c6d1cb f069dfc 1654955 392e60f 1654955 392e60f 1654955 392e60f 1654955 aa29cac 1654955 9f3acc0 392e60f 9f3acc0 392e60f 5c6d1cb 1654955 aa29cac 1654955 0332541 392e60f 5c6d1cb 1654955 0332541 392e60f 9f3acc0 392e60f 0332541 9f3acc0 0332541 1654955 aa29cac 1654955 9f3acc0 0f5dd41 5c6d1cb 392e60f 9f3acc0 392e60f 9f3acc0 392e60f 9f3acc0 392e60f 9f3acc0 392e60f 9f3acc0 392e60f 4ed47f2 9f3acc0 4ed47f2 392e60f 0332541 5c6d1cb 0332541 9f3acc0 392e60f 9f3acc0 392e60f 9f3acc0 392e60f 9f3acc0 392e60f 9f3acc0 392e60f 0332541 5c6d1cb 0332541 0f5dd41 5c6d1cb 0f5dd41 5c6d1cb 0f5dd41 5c6d1cb 0f5dd41 5c6d1cb 0f5dd41 5c6d1cb 0f5dd41 5c6d1cb 0f5dd41 5c6d1cb 0f5dd41 5c6d1cb 0f5dd41 5c6d1cb 0f5dd41 5c6d1cb 0f5dd41 392e60f 1654955 9f3acc0 f069dfc 392e60f 433e26f | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 | """LandmarkDiff -- Facial surgery outcome prediction demo (TPS on CPU)."""
from __future__ import annotations
import logging
import time
import traceback
from pathlib import Path
import cv2
import gradio as gr
import numpy as np
from landmarkdiff.conditioning import render_wireframe
from landmarkdiff.landmarks import FaceLandmarks, extract_landmarks
from landmarkdiff.manipulation import PROCEDURE_LANDMARKS, apply_procedure_preset
from landmarkdiff.masking import generate_surgical_mask
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
GITHUB_URL = "https://github.com/dreamlessx/LandmarkDiff-public"
PROCEDURES = list(PROCEDURE_LANDMARKS.keys())
EXAMPLE_DIR = Path(__file__).parent / "examples"
EXAMPLE_IMAGES = sorted(EXAMPLE_DIR.glob("*.png")) if EXAMPLE_DIR.exists() else []
PROCEDURE_INFO = {
"rhinoplasty": "Nose reshaping (bridge, tip, alar width)",
"blepharoplasty": "Eyelid surgery (lid position, canthal tilt)",
"rhytidectomy": "Facelift (midface, jawline tightening)",
"orthognathic": "Jaw surgery (maxilla/mandible repositioning)",
"brow_lift": "Brow elevation, forehead ptosis reduction",
"mentoplasty": "Chin surgery (projection, vertical height)",
}
# ---------------------------------------------------------------------------
# Bilateral symmetry landmark pairs (MediaPipe face mesh indices)
# ---------------------------------------------------------------------------
SYMMETRY_PAIRS: dict[str, list[tuple[int, int]]] = {
"eyes": [
(33, 263),
(133, 362),
(159, 386),
(145, 374),
],
"brows": [
(70, 300),
(63, 293),
(105, 334),
(66, 296),
(107, 336),
],
"cheeks": [
(116, 345),
(123, 352),
(147, 376),
(187, 411),
(205, 425),
],
"mouth": [
(61, 291),
(78, 308),
(95, 324),
],
"jaw": [
(172, 397),
(136, 365),
(150, 379),
(149, 378),
(176, 400),
],
}
# Midline landmarks: forehead top and chin bottom
MIDLINE_TOP = 10
MIDLINE_BOTTOM = 152
# ---------------------------------------------------------------------------
# Image preprocessing helpers
# ---------------------------------------------------------------------------
def _normalize_to_bgr(image: np.ndarray) -> np.ndarray:
"""Convert any input image format (RGBA, grayscale, etc.) to BGR uint8."""
if image is None:
raise ValueError("No image provided")
img = np.asarray(image)
# Handle float images (0-1 range)
if img.dtype in (np.float32, np.float64):
img = (np.clip(img, 0.0, 1.0) * 255).astype(np.uint8)
# Ensure uint8
if img.dtype != np.uint8:
img = img.astype(np.uint8)
if img.ndim == 2:
# Grayscale -> BGR
return cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)
elif img.ndim == 3:
channels = img.shape[2]
if channels == 4:
# RGBA -> BGR (drop alpha)
return cv2.cvtColor(img, cv2.COLOR_RGBA2BGR)
elif channels == 3:
# RGB -> BGR (Gradio sends RGB)
return cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
elif channels == 1:
return cv2.cvtColor(img.squeeze(-1), cv2.COLOR_GRAY2BGR)
raise ValueError(f"Unsupported image shape: {img.shape}")
def _auto_adjust_brightness(image_bgr: np.ndarray) -> np.ndarray:
"""Auto-adjust brightness/contrast if the image is too dark or washed out.
Uses CLAHE on the L channel of LAB color space for adaptive histogram
equalization that preserves color balance.
"""
lab = cv2.cvtColor(image_bgr, cv2.COLOR_BGR2LAB)
l_channel = lab[:, :, 0]
mean_l = float(np.mean(l_channel))
# Only adjust if clearly too dark (<60) or washed out (>200)
if mean_l < 60 or mean_l > 200:
clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))
lab[:, :, 0] = clahe.apply(l_channel)
return cv2.cvtColor(lab, cv2.COLOR_LAB2BGR)
return image_bgr
def _prepare_image(image_rgb: np.ndarray, size: int = 512) -> tuple[np.ndarray, np.ndarray]:
"""Full preprocessing pipeline: normalize, resize, auto-adjust.
Returns (image_bgr_512, image_rgb_512).
"""
image_bgr = _normalize_to_bgr(image_rgb)
image_bgr = resize_preserve_aspect(image_bgr, size)
image_bgr = _auto_adjust_brightness(image_bgr)
image_rgb_out = cv2.cvtColor(image_bgr, cv2.COLOR_BGR2RGB)
return image_bgr, image_rgb_out
def _detect_face_with_hints(image_bgr: np.ndarray) -> tuple[FaceLandmarks | None, str]:
"""Extract landmarks with better error messages for common failure modes.
Returns (face_or_None, error_hint_string).
"""
try:
face = extract_landmarks(image_bgr)
except Exception as exc:
logger.error("Landmark extraction failed: %s\n%s", exc, traceback.format_exc())
return None, f"Landmark extraction error: {exc}"
if face is not None:
return face, ""
# Try to give a more useful hint about why detection failed.
gray = cv2.cvtColor(image_bgr, cv2.COLOR_BGR2GRAY)
h, w = gray.shape[:2]
# Check if image is mostly black (too dark even after auto-adjust)
if float(np.mean(gray)) < 30:
return None, (
"No face detected -- the image appears very dark. Try a photo with better lighting."
)
# Check for very low contrast (washed out)
if float(np.std(gray)) < 15:
return None, (
"No face detected -- the image has very low contrast. "
"Try a photo with more natural lighting."
)
# Check aspect ratio -- extremely tall/wide may indicate a side profile crop
aspect = w / max(h, 1)
if aspect > 2.5 or aspect < 0.4:
return None, (
"No face detected -- unusual aspect ratio. Use a standard portrait or headshot photo."
)
return None, (
"No face detected. Make sure the photo shows a clear, "
"well-lit frontal face. Side profiles and heavily occluded "
"faces are not supported."
)
# ---------------------------------------------------------------------------
# Symmetry analysis
# ---------------------------------------------------------------------------
def compute_symmetry_score(
face: FaceLandmarks,
) -> tuple[float, dict[str, float]]:
"""Compute bilateral facial symmetry from a FaceLandmarks object.
Reflects left-side landmarks across the facial midline and measures the
Euclidean distance to their right-side counterparts. Distances are
normalized by the inter-pupil distance to make the score scale-invariant.
Args:
face: FaceLandmarks with .pixel_coords property returning (478, 2).
Returns:
(overall_score, region_scores) where scores are 0-100
(100 = perfectly symmetric).
"""
coords = face.pixel_coords # (478, 2) -- property, not method
# Compute facial midline from forehead top (10) and chin bottom (152)
mid_top = coords[MIDLINE_TOP] # (2,)
mid_bot = coords[MIDLINE_BOTTOM] # (2,)
# Midline direction vector and unit normal
midline_dir = mid_bot - mid_top
midline_len = np.linalg.norm(midline_dir)
if midline_len < 1e-6:
# Degenerate case -- landmarks are stacked
return 0.0, {region: 0.0 for region in SYMMETRY_PAIRS}
midline_unit = midline_dir / midline_len
# Normal to midline (pointing right)
midline_normal = np.array([midline_unit[1], -midline_unit[0]])
# Normalization factor: use inter-eye distance (outer corners 33 <-> 263)
# for scale-invariant scoring
inter_eye = float(np.linalg.norm(coords[33] - coords[263]))
if inter_eye < 1e-6:
inter_eye = midline_len * 0.4 # fallback
region_scores: dict[str, float] = {}
all_distances: list[float] = []
for region, pairs in SYMMETRY_PAIRS.items():
region_dists: list[float] = []
for left_idx, right_idx in pairs:
left_pt = coords[left_idx]
right_pt = coords[right_idx]
# Reflect left point across the midline:
# 1. Vector from midline top to the point
v = left_pt - mid_top
# 2. Component along the midline normal
normal_component = np.dot(v, midline_normal)
# 3. Reflected point: subtract twice the normal component
reflected = left_pt - 2.0 * normal_component * midline_normal
# Distance between reflected-left and actual-right
dist = float(np.linalg.norm(reflected - right_pt))
region_dists.append(dist)
# Normalize by inter-eye distance and convert to 0-100 score
if region_dists:
mean_dist = float(np.mean(region_dists))
# Normalized distance as fraction of inter-eye distance
norm_dist = mean_dist / inter_eye
# Convert to score: 0 distance = 100, large distance = 0
# Use exponential decay so small asymmetries are penalized gently
score = 100.0 * np.exp(-3.0 * norm_dist)
region_scores[region] = round(max(0.0, min(100.0, score)), 1)
all_distances.extend(region_dists)
else:
region_scores[region] = 0.0
# Overall score: weighted mean of all pair distances
if all_distances:
overall_norm = float(np.mean(all_distances)) / inter_eye
overall = 100.0 * np.exp(-3.0 * overall_norm)
overall = round(max(0.0, min(100.0, overall)), 1)
else:
overall = 0.0
return overall, region_scores
def render_symmetry_overlay(
image_bgr: np.ndarray,
face: FaceLandmarks,
region_scores: dict[str, float],
) -> np.ndarray:
"""Draw a symmetry visualization overlay on the image.
Draws the facial midline and color-codes bilateral landmark pairs by
their region symmetry score: green (>80), yellow (50-80), red (<50).
"""
canvas = image_bgr.copy()
coords = face.pixel_coords
# Draw midline
mid_top = coords[MIDLINE_TOP].astype(int)
mid_bot = coords[MIDLINE_BOTTOM].astype(int)
cv2.line(canvas, tuple(mid_top), tuple(mid_bot), (255, 200, 0), 2, cv2.LINE_AA)
# Small label at midline top
cv2.putText(
canvas,
"midline",
(int(mid_top[0]) + 5, int(mid_top[1]) - 5),
cv2.FONT_HERSHEY_SIMPLEX,
0.4,
(255, 200, 0),
1,
cv2.LINE_AA,
)
def _score_color(score: float) -> tuple[int, int, int]:
"""BGR color based on symmetry score."""
if score >= 80:
return (0, 200, 0) # green
elif score >= 50:
return (0, 200, 220) # yellow (BGR)
else:
return (0, 0, 220) # red
for region, pairs in SYMMETRY_PAIRS.items():
score = region_scores.get(region, 0.0)
color = _score_color(score)
for left_idx, right_idx in pairs:
lp = coords[left_idx].astype(int)
rp = coords[right_idx].astype(int)
# Draw landmark dots
cv2.circle(canvas, tuple(lp), 3, color, -1, cv2.LINE_AA)
cv2.circle(canvas, tuple(rp), 3, color, -1, cv2.LINE_AA)
# Draw thin connecting line across midline
cv2.line(canvas, tuple(lp), tuple(rp), color, 1, cv2.LINE_AA)
# Draw region labels with scores
# Position labels near each region's centroid
region_label_offsets: dict[str, tuple[int, int]] = {
"eyes": (0, -15),
"brows": (0, -10),
"cheeks": (15, 0),
"mouth": (0, 10),
"jaw": (0, 15),
}
for region, pairs in SYMMETRY_PAIRS.items():
score = region_scores.get(region, 0.0)
color = _score_color(score)
# Compute centroid of the region landmarks
region_pts = []
for left_idx, right_idx in pairs:
region_pts.append(coords[left_idx])
region_pts.append(coords[right_idx])
centroid = np.mean(region_pts, axis=0).astype(int)
ox, oy = region_label_offsets.get(region, (0, 0))
label_pos = (int(centroid[0]) + ox, int(centroid[1]) + oy)
label = f"{region}: {score:.0f}"
cv2.putText(
canvas,
label,
label_pos,
cv2.FONT_HERSHEY_SIMPLEX,
0.4,
color,
1,
cv2.LINE_AA,
)
return canvas
def _format_symmetry_text(
overall: float,
region_scores: dict[str, float],
prefix: str = "",
) -> str:
"""Format symmetry scores into a readable text block."""
lines = []
if prefix:
lines.append(prefix)
lines.append(f"Overall symmetry: {overall:.1f}/100")
for region, score in region_scores.items():
bar_len = int(score / 5)
bar = "|" * bar_len + "." * (20 - bar_len)
lines.append(f" {region:>6s}: {score:5.1f} [{bar}]")
return "\n".join(lines)
# ---------------------------------------------------------------------------
# Symmetry tab callbacks
# ---------------------------------------------------------------------------
def analyze_symmetry(image_rgb: np.ndarray):
"""Analyze facial symmetry from an uploaded photo."""
if image_rgb is None:
b = _blank()
return b, "Upload a face photo to analyze symmetry."
try:
image_bgr, image_rgb_512 = _prepare_image(image_rgb, 512)
except Exception as exc:
logger.error("Image conversion failed: %s", exc)
b = _blank()
return b, f"Image conversion failed: {exc}"
face, hint = _detect_face_with_hints(image_bgr)
if face is None:
return image_rgb_512, hint
overall, region_scores = compute_symmetry_score(face)
overlay_bgr = render_symmetry_overlay(image_bgr, face, region_scores)
overlay_rgb = cv2.cvtColor(overlay_bgr, cv2.COLOR_BGR2RGB)
text = _format_symmetry_text(overall, region_scores)
return overlay_rgb, text
def analyze_symmetry_comparison(
pre_image_rgb: np.ndarray,
post_image_rgb: np.ndarray,
):
"""Compare symmetry between pre- and post-procedure photos."""
b = _blank()
empty = (b, b, "Upload both pre and post photos to compare.")
if pre_image_rgb is None or post_image_rgb is None:
return empty
try:
pre_bgr, _ = _prepare_image(pre_image_rgb, 512)
post_bgr, _ = _prepare_image(post_image_rgb, 512)
except Exception as exc:
logger.error("Image conversion failed: %s", exc)
return b, b, f"Image conversion failed: {exc}"
pre_face, pre_hint = _detect_face_with_hints(pre_bgr)
if pre_face is None:
return b, b, f"Pre-procedure: {pre_hint}"
post_face, post_hint = _detect_face_with_hints(post_bgr)
if post_face is None:
return b, b, f"Post-procedure: {post_hint}"
pre_overall, pre_regions = compute_symmetry_score(pre_face)
post_overall, post_regions = compute_symmetry_score(post_face)
pre_overlay = render_symmetry_overlay(pre_bgr, pre_face, pre_regions)
post_overlay = render_symmetry_overlay(post_bgr, post_face, post_regions)
pre_rgb = cv2.cvtColor(pre_overlay, cv2.COLOR_BGR2RGB)
post_rgb = cv2.cvtColor(post_overlay, cv2.COLOR_BGR2RGB)
lines = []
lines.append(_format_symmetry_text(pre_overall, pre_regions, prefix="-- Pre-procedure --"))
lines.append("")
lines.append(_format_symmetry_text(post_overall, post_regions, prefix="-- Post-procedure --"))
lines.append("")
delta = post_overall - pre_overall
direction = "improved" if delta > 0 else "decreased"
lines.append(f"Change: {delta:+.1f} ({direction})")
# Per-region deltas
for region in pre_regions:
d = post_regions.get(region, 0.0) - pre_regions[region]
lines.append(f" {region:>6s}: {d:+.1f}")
return pre_rgb, post_rgb, "\n".join(lines)
# ---------------------------------------------------------------------------
# Core pipeline functions
# ---------------------------------------------------------------------------
def warp_image_tps(image, src_pts, dst_pts):
"""Thin-plate spline warp (CPU only)."""
from landmarkdiff.synthetic.tps_warp import warp_image_tps as _warp
return _warp(image, src_pts, dst_pts)
def resize_preserve_aspect(image, size=512):
"""Resize to square canvas, padding to preserve aspect ratio."""
h, w = image.shape[:2]
scale = size / max(h, w)
new_w, new_h = int(w * scale), int(h * scale)
resized = cv2.resize(image, (new_w, new_h), interpolation=cv2.INTER_LANCZOS4)
canvas = np.zeros((size, size, 3), dtype=np.uint8)
y_off = (size - new_h) // 2
x_off = (size - new_w) // 2
canvas[y_off : y_off + new_h, x_off : x_off + new_w] = resized
return canvas
def mask_composite(warped, original, mask):
"""Alpha-blend warped region into original using mask."""
mask_3 = np.stack([mask] * 3, axis=-1) if mask.ndim == 2 else mask
return (warped * mask_3 + original * (1.0 - mask_3)).astype(np.uint8)
def _blank():
return np.zeros((512, 512, 3), dtype=np.uint8)
def process_image(image_rgb, procedure, intensity):
"""Run the TPS pipeline on a single image, including symmetry scores."""
if image_rgb is None:
b = _blank()
return b, b, b, b, "Upload a face photo to begin."
t0 = time.monotonic()
try:
image_bgr, image_rgb_512 = _prepare_image(image_rgb, 512)
except Exception as exc:
logger.error("Image conversion failed: %s", exc)
b = _blank()
return b, b, b, b, f"Image conversion failed: {exc}"
face, hint = _detect_face_with_hints(image_bgr)
if face is None:
if hint:
return image_rgb_512, image_rgb_512, image_rgb_512, image_rgb_512, hint
return (
image_rgb_512,
image_rgb_512,
image_rgb_512,
image_rgb_512,
"No face detected. Try a clearer, well-lit frontal photo.",
)
try:
manipulated = apply_procedure_preset(face, procedure, float(intensity), image_size=512)
wireframe = render_wireframe(manipulated, width=512, height=512)
wireframe_rgb = cv2.cvtColor(wireframe, cv2.COLOR_GRAY2RGB)
mask = generate_surgical_mask(face, procedure, 512, 512)
mask_vis = cv2.cvtColor((mask * 255).astype(np.uint8), cv2.COLOR_GRAY2RGB)
warped = warp_image_tps(image_bgr, face.pixel_coords, manipulated.pixel_coords)
composited = mask_composite(warped, image_bgr, mask)
composited_rgb = cv2.cvtColor(composited, cv2.COLOR_BGR2RGB)
displacement = np.mean(np.linalg.norm(manipulated.pixel_coords - face.pixel_coords, axis=1))
elapsed = time.monotonic() - t0
# Compute symmetry for original and predicted result
pre_overall, pre_regions = compute_symmetry_score(face)
post_overall, post_regions = compute_symmetry_score(manipulated)
sym_delta = post_overall - pre_overall
sym_arrow = "+" if sym_delta > 0 else ""
info_lines = [
"--- Procedure ---",
f" Type: {procedure.replace('_', ' ').title()}",
f" Intensity: {intensity:.0f}%",
f" Description: {PROCEDURE_INFO.get(procedure, '')}",
"",
"--- Detection ---",
f" Landmarks: {len(face.landmarks)} points",
f" Confidence: {face.confidence:.2f}",
f" Avg shift: {displacement:.1f} px",
"",
"--- Symmetry ---",
f" Before: {pre_overall:.1f} / 100",
f" After: {post_overall:.1f} / 100",
f" Change: {sym_arrow}{sym_delta:.1f}",
"",
"--- Performance ---",
f" Time: {elapsed:.2f}s",
" Mode: TPS (CPU)",
]
info = "\n".join(info_lines)
return wireframe_rgb, mask_vis, composited_rgb, image_rgb_512, info
except Exception as exc:
logger.error("Processing failed: %s\n%s", exc, traceback.format_exc())
b = _blank()
return b, b, b, b, f"Processing error: {exc}"
def compare_procedures(image_rgb, intensity):
"""Compare all six procedures at the same intensity."""
if image_rgb is None:
return [_blank()] * len(PROCEDURES)
try:
image_bgr, _ = _prepare_image(image_rgb, 512)
face, _ = _detect_face_with_hints(image_bgr)
if face is None:
rgb = cv2.cvtColor(image_bgr, cv2.COLOR_BGR2RGB)
return [rgb] * len(PROCEDURES)
results = []
for proc in PROCEDURES:
manip = apply_procedure_preset(face, proc, float(intensity), image_size=512)
mask = generate_surgical_mask(face, proc, 512, 512)
warped = warp_image_tps(image_bgr, face.pixel_coords, manip.pixel_coords)
comp = mask_composite(warped, image_bgr, mask)
results.append(cv2.cvtColor(comp, cv2.COLOR_BGR2RGB))
return results
except Exception as exc:
logger.error("Compare failed: %s\n%s", exc, traceback.format_exc())
return [_blank()] * len(PROCEDURES)
def intensity_sweep(image_rgb, procedure):
"""Generate results at 0%, 20%, 40%, 60%, 80%, 100% intensity."""
if image_rgb is None:
return []
try:
image_bgr, _ = _prepare_image(image_rgb, 512)
face, _ = _detect_face_with_hints(image_bgr)
if face is None:
return []
results = []
for val in [0, 20, 40, 60, 80, 100]:
if val == 0:
results.append((cv2.cvtColor(image_bgr, cv2.COLOR_BGR2RGB), "0%"))
continue
manip = apply_procedure_preset(face, procedure, float(val), image_size=512)
mask = generate_surgical_mask(face, procedure, 512, 512)
warped = warp_image_tps(image_bgr, face.pixel_coords, manip.pixel_coords)
comp = mask_composite(warped, image_bgr, mask)
results.append((cv2.cvtColor(comp, cv2.COLOR_BGR2RGB), f"{val}%"))
return results
except Exception as exc:
logger.error("Sweep failed: %s\n%s", exc, traceback.format_exc())
return []
# ---------------------------------------------------------------------------
# Build the Gradio UI
# ---------------------------------------------------------------------------
_proc_table = "\n".join(
f"| {name.replace('_', ' ').title()} | {desc} |" for name, desc in PROCEDURE_INFO.items()
)
_CSS = """
.header-banner {
background: linear-gradient(135deg, #1a1a2e 0%, #16213e 50%, #0f3460 100%);
border-radius: 12px;
padding: 24px 32px;
margin-bottom: 8px;
color: white;
}
.header-banner h1 { color: white !important; margin-bottom: 4px; font-size: 2em; }
.header-banner p { color: #ccd; margin: 4px 0; font-size: 0.95em; }
.header-banner a { color: #7eb8f7; text-decoration: none; }
.header-banner a:hover { text-decoration: underline; }
.link-bar { display: flex; gap: 16px; margin-top: 10px; flex-wrap: wrap; }
.info-output textarea {
font-family: 'SF Mono', 'Fira Code', 'Consolas', monospace !important;
font-size: 0.88em !important;
line-height: 1.6 !important;
}
"""
with gr.Blocks(
title="LandmarkDiff -- Facial Surgery Prediction",
theme=gr.themes.Soft(),
css=_CSS,
) as demo:
gr.HTML(
f"""<div class="header-banner">
<h1>LandmarkDiff</h1>
<p>
Anatomically-conditioned facial surgery outcome prediction from standard clinical
photography. Upload a face photo, select a procedure, adjust intensity, and see
the predicted result in real time.
</p>
<p style="font-size:0.85em; color:#aab;">
Powered by MediaPipe 478-point face mesh, thin-plate spline warping, and
procedure-specific anatomical displacement models. Runs entirely on CPU.
This 2D demo is the foundation -- 3D face reconstruction from phone video
is on the roadmap.
</p>
<div class="link-bar">
<a href="{GITHUB_URL}">GitHub</a>
<a href="{GITHUB_URL}/tree/main/docs">Documentation</a>
<a href="{GITHUB_URL}/wiki">Wiki</a>
<a href="{GITHUB_URL}/discussions">Discussions</a>
</div>
</div>"""
)
# -- Tab 1: Single Procedure --
with gr.Tab("Single Procedure"):
with gr.Row():
with gr.Column(scale=1):
input_image = gr.Image(label="Face Photo", type="numpy", height=350)
procedure = gr.Radio(
choices=PROCEDURES,
value="rhinoplasty",
label="Procedure",
info="Select a surgical procedure to simulate",
)
# Show a brief description for each procedure
_proc_desc_md = " | ".join(
f"**{k.replace('_', ' ').title()}**: {v}" for k, v in PROCEDURE_INFO.items()
)
gr.Markdown(
f"<div style='font-size:0.82em;color:#666;line-height:1.5;'>"
f"{_proc_desc_md}</div>"
)
intensity = gr.Slider(
0,
100,
50,
step=1,
label="Intensity (%)",
info="0 = no change, 100 = maximum effect",
)
run_btn = gr.Button("Generate Prediction", variant="primary", size="lg")
info_box = gr.Textbox(
label="Results",
lines=10,
interactive=False,
elem_classes=["info-output"],
)
with gr.Column(scale=2):
with gr.Row():
out_wireframe = gr.Image(label="Deformed Wireframe", height=256)
out_mask = gr.Image(label="Surgical Mask", height=256)
with gr.Row():
out_result = gr.Image(label="Predicted Result", height=256)
out_original = gr.Image(label="Original", height=256)
if EXAMPLE_IMAGES:
gr.Examples(
examples=[[str(p)] for p in EXAMPLE_IMAGES],
inputs=[input_image],
label="Example faces (click to load)",
)
with gr.Accordion("Photo Tips for Best Results", open=False):
gr.Markdown(
"- **Front-facing**: Use a straight-on frontal photo, "
"not a side profile\n"
"- **Good lighting**: Even, natural lighting works best. "
"Avoid harsh shadows\n"
"- **Neutral expression**: Keep a relaxed, neutral face "
"for accurate landmark detection\n"
"- **No obstructions**: Remove glasses, hats, or anything "
"covering the face\n"
"- **Resolution**: At least 256x256 pixels. The image will "
"be resized to 512x512 internally\n"
"- **Single face**: Make sure only one face is clearly "
"visible in the frame"
)
outputs = [out_wireframe, out_mask, out_result, out_original, info_box]
_inputs = [input_image, procedure, intensity]
run_btn.click(fn=process_image, inputs=_inputs, outputs=outputs)
# Auto-trigger on image upload and procedure change, but not on every
# slider tick during drag (each tick would re-run TPS on free CPU,
# causing severe lag). Use .release so it fires once on mouse-up.
input_image.change(fn=process_image, inputs=_inputs, outputs=outputs)
procedure.change(fn=process_image, inputs=_inputs, outputs=outputs)
intensity.release(fn=process_image, inputs=_inputs, outputs=outputs)
# -- Tab 2: Compare Procedures --
with gr.Tab("Compare All"):
gr.Markdown("All six procedures at the same intensity, side by side.")
with gr.Row():
with gr.Column(scale=1):
cmp_image = gr.Image(label="Face Photo", type="numpy", height=300)
cmp_intensity = gr.Slider(0, 100, 50, step=1, label="Intensity (%)")
cmp_btn = gr.Button("Compare", variant="primary", size="lg")
with gr.Column(scale=2):
cmp_outputs = []
for row_idx in range(2):
with gr.Row():
for col_idx in range(3):
idx = row_idx * 3 + col_idx
if idx < len(PROCEDURES):
cmp_outputs.append(
gr.Image(
label=PROCEDURES[idx].replace("_", " ").title(),
height=200,
)
)
if EXAMPLE_IMAGES:
gr.Examples(
examples=[[str(p)] for p in EXAMPLE_IMAGES],
inputs=[cmp_image],
label="Examples",
)
cmp_btn.click(fn=compare_procedures, inputs=[cmp_image, cmp_intensity], outputs=cmp_outputs)
# -- Tab 3: Intensity Sweep --
with gr.Tab("Intensity Sweep"):
gr.Markdown("See a procedure at 0% through 100% in six steps.")
with gr.Row():
with gr.Column(scale=1):
sweep_image = gr.Image(label="Face Photo", type="numpy", height=300)
sweep_proc = gr.Radio(choices=PROCEDURES, value="rhinoplasty", label="Procedure")
sweep_btn = gr.Button("Sweep", variant="primary", size="lg")
with gr.Column(scale=2):
sweep_gallery = gr.Gallery(label="0% to 100%", columns=3, height=400)
if EXAMPLE_IMAGES:
gr.Examples(
examples=[[str(p)] for p in EXAMPLE_IMAGES],
inputs=[sweep_image],
label="Examples",
)
sweep_btn.click(
fn=intensity_sweep,
inputs=[sweep_image, sweep_proc],
outputs=[sweep_gallery],
)
# -- Tab 4: Symmetry Analysis --
with gr.Tab("Symmetry Analysis"):
gr.Markdown(
"### Bilateral Facial Symmetry\n\n"
"Analyzes left-right symmetry across five facial regions "
"(eyes, brows, cheeks, mouth, jaw) using MediaPipe 478-point "
"face mesh landmark pairs. The midline is computed from the "
"forehead apex to the chin, and left landmarks are reflected "
"across it to measure deviation from the right side.\n\n"
"**Score interpretation:** 90-100 = highly symmetric, "
"70-89 = mild asymmetry, <70 = notable asymmetry."
)
with gr.Tabs():
# Sub-tab: Single image analysis
with gr.Tab("Single Photo"):
with gr.Row():
with gr.Column(scale=1):
sym_image = gr.Image(
label="Face Photo",
type="numpy",
height=350,
)
sym_btn = gr.Button(
"Analyze Symmetry",
variant="primary",
size="lg",
)
with gr.Column(scale=1):
sym_overlay = gr.Image(label="Symmetry Overlay", height=350)
sym_scores_box = gr.Textbox(
label="Symmetry Scores",
lines=8,
interactive=False,
)
if EXAMPLE_IMAGES:
gr.Examples(
examples=[[str(p)] for p in EXAMPLE_IMAGES],
inputs=[sym_image],
label="Examples",
)
sym_btn.click(
fn=analyze_symmetry,
inputs=[sym_image],
outputs=[sym_overlay, sym_scores_box],
)
# Sub-tab: Pre vs post comparison
with gr.Tab("Pre vs Post Comparison"):
gr.Markdown(
"Upload a pre-procedure and post-procedure photo to compare "
"how symmetry changed."
)
with gr.Row():
sym_pre_image = gr.Image(
label="Pre-Procedure",
type="numpy",
height=300,
)
sym_post_image = gr.Image(
label="Post-Procedure",
type="numpy",
height=300,
)
sym_cmp_btn = gr.Button(
"Compare Symmetry",
variant="primary",
size="lg",
)
with gr.Row():
sym_pre_overlay = gr.Image(
label="Pre Symmetry Overlay",
height=300,
)
sym_post_overlay = gr.Image(
label="Post Symmetry Overlay",
height=300,
)
sym_cmp_box = gr.Textbox(
label="Comparison",
lines=14,
interactive=False,
)
sym_cmp_btn.click(
fn=analyze_symmetry_comparison,
inputs=[sym_pre_image, sym_post_image],
outputs=[sym_pre_overlay, sym_post_overlay, sym_cmp_box],
)
gr.HTML(
f"<div style='text-align:center;color:#888;font-size:0.78em;padding:12px 8px;"
f"border-top:1px solid #eee;margin-top:12px;'>"
f"LandmarkDiff v0.2.2 · TPS on CPU · "
f"MediaPipe 478-point mesh · "
f"<a href='{GITHUB_URL}' style='color:#7eb8f7;'>GitHub</a> · "
f"MIT License · For research and educational purposes only"
f"</div>"
)
if __name__ == "__main__":
demo.launch(show_error=True)
|