Spaces:
Sleeping
Sleeping
File size: 32,122 Bytes
d5e4d53 8776d83 d5e4d53 8776d83 d5e4d53 8776d83 d5e4d53 8776d83 d5e4d53 8776d83 d5e4d53 581240c 8776d83 581240c 8776d83 581240c 8776d83 d5e4d53 581240c d5e4d53 8776d83 d5e4d53 581240c d5e4d53 581240c d5e4d53 581240c d5e4d53 581240c d5e4d53 8776d83 d5e4d53 8776d83 581240c 8776d83 581240c 8776d83 581240c d5e4d53 8776d83 d5e4d53 8776d83 d5e4d53 8776d83 d5e4d53 8776d83 581240c 8776d83 d5e4d53 8776d83 d5e4d53 8776d83 d5e4d53 8776d83 d5e4d53 8776d83 d5e4d53 581240c 8776d83 581240c 8776d83 d5e4d53 8776d83 581240c 8776d83 d5e4d53 581240c d5e4d53 581240c d5e4d53 3b5257a 581240c 3b5257a 581240c d5e4d53 581240c d5e4d53 8776d83 d5e4d53 8776d83 d5e4d53 8776d83 d5e4d53 8776d83 d5e4d53 8776d83 d5e4d53 8776d83 d5e4d53 8776d83 d5e4d53 8776d83 d5e4d53 3b5257a 581240c 3b5257a 581240c 3b5257a d5e4d53 8776d83 d5e4d53 581240c d5e4d53 581240c 3b5257a 581240c 3b5257a d5e4d53 581240c 8776d83 d5e4d53 3b5257a d5e4d53 581240c 3b5257a 581240c d5e4d53 3b5257a 581240c 3b5257a d5e4d53 581240c 8776d83 d5e4d53 8776d83 581240c d5e4d53 581240c d5e4d53 581240c d5e4d53 581240c d5e4d53 581240c d5e4d53 581240c 8776d83 d5e4d53 3b5257a 581240c d5e4d53 8776d83 581240c 8776d83 581240c d5e4d53 581240c 8776d83 d5e4d53 8776d83 d5e4d53 8776d83 d5e4d53 8776d83 d5e4d53 8776d83 d5e4d53 8776d83 d5e4d53 8776d83 d5e4d53 8776d83 d5e4d53 8776d83 d5e4d53 8776d83 581240c 3b5257a 581240c 3b5257a 581240c 8776d83 d5e4d53 8776d83 3b5257a 8776d83 d5e4d53 8776d83 d5e4d53 8776d83 d5e4d53 8776d83 3b5257a 581240c 8776d83 581240c 8776d83 581240c 3b5257a 581240c 8776d83 3b5257a d5e4d53 581240c 8776d83 d5e4d53 8776d83 d5e4d53 581240c 8776d83 d5e4d53 581240c 3b5257a d5e4d53 581240c 8776d83 3b5257a 581240c 8776d83 581240c d5e4d53 8776d83 d5e4d53 8776d83 d5e4d53 8776d83 d5e4d53 8776d83 581240c 8776d83 581240c 8776d83 581240c 8776d83 581240c 8776d83 581240c 8776d83 d5e4d53 8776d83 581240c 8776d83 581240c d5e4d53 121dec8 d5e4d53 8776d83 d5e4d53 581240c 8776d83 3b5257a 581240c d5e4d53 581240c d5e4d53 581240c d5e4d53 3b5257a d5e4d53 3b5257a d5e4d53 8776d83 581240c d5e4d53 581240c d5e4d53 8776d83 581240c d5e4d53 581240c 8776d83 d5e4d53 3b5257a 581240c d5e4d53 581240c d5e4d53 581240c d5e4d53 581240c | 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 | import os, warnings, time, glob, tempfile, threading
warnings.filterwarnings("ignore")
os.environ["GRADIO_ANALYTICS_ENABLED"] = "False"
from pathlib import Path
import numpy as np
from PIL import Image, ImageDraw, ImageOps
import gradio as gr
print("\n" + "="*60)
print("🔍 INTERACTIVE-MEN-RT DEMO DEBUG INFO")
print("="*60)
print(f"📦 Gradio version: {gr.__version__}")
print(f"📁 Current directory: {os.getcwd()}")
print(f"📁 Directory contents: {os.listdir('.')}")
DATA_ROOT = Path("./samples")
print(f"\n📂 DATA_ROOT: {DATA_ROOT}")
print(f"📂 DATA_ROOT exists: {DATA_ROOT.exists()}")
if DATA_ROOT.exists():
print(f"📂 DATA_ROOT contents: {list(DATA_ROOT.iterdir())}")
EXAMPLES_CHECK = ["BraTS-MEN-RT-0071-1"]
for case in EXAMPLES_CHECK:
case_dir = DATA_ROOT / case
print(f"\n 📦 Case: {case}")
print(f" Exists: {case_dir.exists()}")
if case_dir.exists():
files = list(case_dir.iterdir())
print(f" Files: {[f.name for f in files]}")
has_t1c = any(f.name.endswith('_t1c.nii.gz') for f in files)
print(f" ✓ Has T1c: {has_t1c}")
else:
print("❌ samples folder NOT FOUND!")
print("\n" + "="*60 + "\n")
# ---------- optional deps ----------
try:
import nibabel as nib
HAVE_NIB = True
print("✓ nibabel available")
except Exception as e:
HAVE_NIB = False
print(f"✗ nibabel not available: {e}")
try:
from scipy import ndimage as ndi
HAVE_SCIPY = True
print("✓ scipy available")
except Exception as e:
HAVE_SCIPY = False
print(f"✗ scipy not available: {e}")
# ---------- model predictor ----------
PREDICTOR = None
DEVICE = "cuda:0"
from huggingface_hub import snapshot_download
try:
_repo_root = snapshot_download("hanjang/Interactive-MEN-RT",
allow_patterns=["nnUNetInteractionTrainer__nnUNetPlans__3d_fullres_scratch/**"])
CKPT = os.path.join(_repo_root, "nnUNetInteractionTrainer__nnUNetPlans__3d_fullres_scratch")
print(f"[INFO] Checkpoint path: {CKPT}")
if os.path.exists(CKPT):
contents = os.listdir(CKPT)
print(f"[INFO] Checkpoint contents: {contents}")
fold_0 = os.path.join(CKPT, "fold_0")
if os.path.exists(fold_0):
print(f"[INFO] fold_0 contents: {os.listdir(fold_0)}")
else:
print(f"[ERROR] Checkpoint path does not exist!")
CKPT = None
except Exception as e:
print(f"[ERROR] Failed to download checkpoint: {e}")
CKPT = None
for env in ("nnUNet_raw", "nnUNet_preprocessed", "nnUNet_results"):
os.environ.setdefault(env, tempfile.mkdtemp(prefix=f"{env}_"))
def _init_predictor_once():
global PREDICTOR
if PREDICTOR is not None:
return True
if CKPT is None:
print("[WARN] No checkpoint available, will use fallback only")
return False
try:
import torch
from Interactive_MEN_RT_predictor import InteractiveMENRTPredictor
dev = torch.device(DEVICE if torch.cuda.is_available() else "cpu")
pred = InteractiveMENRTPredictor(
device=dev, use_torch_compile=False, do_autozoom=False, verbose=False
)
pred.initialize_from_trained_model_folder(
model_training_output_dir=CKPT, use_fold=0, checkpoint_name="checkpoint_best.pth"
)
PREDICTOR = pred
try:
if torch.cuda.is_available():
x = np.zeros((1, 8, 8, 8), np.float32)
pred.reset_interactions()
pred.set_image(x)
pred.set_target_buffer(np.zeros_like(x[0], np.float32))
pred._finish_preprocessing_and_initialize_interactions()
torch.cuda.synchronize()
except Exception:
pass
print("[MODEL] ready")
return True
except Exception as e:
print(f"[MODEL] init failed: {e}")
return False
def preload_model_in_background():
threading.Thread(target=_init_predictor_once, daemon=True).start()
# ---------- config ----------
EXAMPLES = ["BraTS-MEN-RT-0071-1"]
RENDER_PX_DEFAULT = 384
ROT_CCW = True
# colors
ACCENT_HEX = "#1e90ff"
CROSS_RGB = (30, 144, 255)
GT_RGBA_FILL = (255, 215, 0, 128)
PR_RGBA_FILL = (255, 60, 60, 128)
SEED_RGB = (89, 224, 154)
BBOX_RGB = (255, 140, 0)
# ---------- state ----------
class State:
def __init__(self):
self.vol=None; self.shape=None
self.gt=None; self.pred=None
self.case_id=None; self.loaded=False
self.cross={"x":0,"y":0,"z":0}
self.slice={"axial":0,"sagittal":0,"coronal":0}
self.seeds=[]
self.seed_views=[]
self.render_px=RENDER_PX_DEFAULT
self.disp_wh={"axial":(RENDER_PX_DEFAULT,RENDER_PX_DEFAULT),
"sagittal":(RENDER_PX_DEFAULT,RENDER_PX_DEFAULT),
"coronal":(RENDER_PX_DEFAULT,RENDER_PX_DEFAULT)}
self.active_view="axial"
self.bbox_mode = False
self.bbox_points = []
self.bboxes = []
self.ref_affine = None
self.ref_header = None
S = State()
# ---------- utils ----------
def _norm01(a):
a=a.astype(np.float32)
p2,p98=np.percentile(a,2),np.percentile(a,98)
if p98<=p2: p2,p98=float(a.min()),float(a.max()) or 1.0
return np.clip((a-p2)/max(p98-p2,1e-6),0,1)
def _resize_slice_nearest(arr2d,w,h):
im=Image.fromarray(arr2d); im=im.resize((w,h),Image.NEAREST); return np.array(im)
def _rot90_if_needed(img_or_np):
if not ROT_CCW: return img_or_np
if isinstance(img_or_np, Image.Image): return img_or_np.rotate(90, expand=True)
return np.rot90(img_or_np, k=1)
# ---------- IO ----------
def _load_png_stack(case_dir):
pngs = sorted(glob.glob(str(case_dir / "png_axial" / "*.png")))
if not pngs:
pngs = sorted(glob.glob(str(case_dir / "png_axial" / "*.jpg")))
if not pngs: return None, None, None
t0=time.time()
arr=[np.array(Image.open(p).convert("L")) for p in pngs]
vol=np.stack(arr,axis=2).astype(np.float32)
vol=_norm01(vol)
print(f"[PIL] {len(pngs)} slices -> {vol.shape} in {time.time()-t0:.2f}s")
return vol, None, None
def _load_nifti(case_dir,case_id,ds=1):
if not HAVE_NIB: return None, None, None
p=case_dir/f"{case_id}_t1c.nii.gz"
if not p.exists(): return None, None, None
t0=time.time()
nii=nib.load(str(p))
arr=np.asanyarray(nii.dataobj[::ds,::ds,::ds],dtype=np.float32)
arr=_norm01(arr)
print(f"[NIfTI] {case_id} -> {arr.shape} in {time.time()-t0:.2f}s")
return arr, nii.affine, nii.header
def _resample_mask_to_vol_shape(mask_xyz, vol_shape_xyz):
mx,my,mz=mask_xyz.shape; vx,vy,vz=vol_shape_xyz
out=np.zeros((vx,vy,vz),dtype=np.uint8)
for k in range(vz):
src_k=int(round(k*(mz-1)/max(vz-1,1)))
sl=(mask_xyz[:,:,src_k]>0).astype(np.uint8)*255
im=Image.fromarray(sl).resize((vy,vx),Image.NEAREST)
out[:,:,k]=(np.array(im)>0).astype(np.uint8)
return out
def _load_gt(case_dir,case_id,vol_shape):
candidates = [
f"{case_id}_gtv.nii.gz", f"{case_id}_seg.nii.gz", f"{case_id}_gt.nii.gz",
"gtv.nii.gz", "seg.nii.gz", "gt.nii.gz",
f"{case_id}_gtv.nii", f"{case_id}_seg.nii", f"{case_id}_gt.nii",
]
if HAVE_NIB:
for name in candidates:
p = case_dir/name
if p.exists():
try:
m=np.asanyarray(nib.load(str(p)).dataobj,dtype=np.uint8)
print(f"[GT] found {p.name} raw={m.shape}")
m=_resample_mask_to_vol_shape(m,vol_shape)
print(f"[GT] resized -> {m.shape}")
return (m>0).astype(np.uint8)
except Exception as e:
print(f"[GT] load error {p.name}: {e}")
print("[GT] not found.")
return None
def load_case(case_id):
case_dir=DATA_ROOT/case_id
vol, affine, header = _load_png_stack(case_dir)
if vol is None:
vol, affine, header = _load_nifti(case_dir,case_id,ds=1)
if vol is None:
Z=96
x=np.linspace(-1,1,RENDER_PX_DEFAULT)[:,None,]
y=np.linspace(-1,1,RENDER_PX_DEFAULT)[None,:,]
z=np.linspace(-1,1,Z)[None,None,:]
vol=np.exp(-(x**2+y**2+z**2)*6).astype(np.float32)
affine = np.eye(4)
header = None
print("[VOL] dummy")
S.vol=vol; S.shape=vol.shape; S.pred=None
S.seeds=[]; S.seed_views=[]
S.bbox_mode=False; S.bbox_points=[]; S.bboxes=[]
S.case_id=case_id
S.ref_affine = affine
S.ref_header = header
X,Y,Z=S.shape
S.cross={"x":X//2,"y":Y//2,"z":Z//2}
S.slice={"sagittal":S.cross["x"],"coronal":S.cross["y"],"axial":S.cross["z"]}
S.gt=_load_gt(case_dir,case_id,S.shape)
S.render_px=RENDER_PX_DEFAULT
S.active_view="axial"
S.loaded=True
print(f"[LOAD] {case_id} | shape={S.shape}")
# ---------- 2D rendering ----------
def _slice2d(view):
if view=="axial": sl=S.vol[:,:,S.slice["axial"]]
elif view=="sagittal": sl=S.vol[S.slice["sagittal"],:,:].T
else: sl=S.vol[:,S.slice["coronal"],:].T
return _rot90_if_needed(sl)
def _cross_pix_on_rot(view,w,h,x=None,y=None,z=None):
X,Y,Z=S.shape
if x is None: x=S.cross["x"]
if y is None: y=S.cross["y"]
if z is None: z=S.cross["z"]
if view=="axial":
u=int(round(x*(w-1)/max(X-1,1)))
v=int(round((Y-1-y)*(h-1)/max(Y-1,1)))
elif view=="sagittal":
u=int(round(z*(w-1)/max(Z-1,1)))
v=int(round((Y-1-y)*(h-1)/max(Y-1,1)))
else:
u=int(round(z*(w-1)/max(Z-1,1)))
v=int(round((X-1-x)*(h-1)/max(X-1,1)))
return u,v
def _draw_cross(img_draw, view, w, h):
u,v=_cross_pix_on_rot(view,w,h)
img_draw.line([(u,0),(u,h)],fill=CROSS_RGB,width=1)
img_draw.line([(0,v),(w,v)],fill=CROSS_RGB,width=1)
img_draw.ellipse((u-5,v-5,u+5,v+5),fill=(255,255,255),outline=ACCENT_HEX,width=2)
def render_top(view):
if not S.loaded: return None
sl=_slice2d(view)
im=Image.fromarray((sl*255).astype(np.uint8)).resize((S.render_px,S.render_px),Image.BILINEAR).convert("RGB")
w=h=S.render_px
dr=ImageDraw.Draw(im)
_draw_cross(dr, view, w, h)
S.disp_wh[view]=im.size
return im
def _axial_mask2d_rot(mask3d):
if mask3d is None: return None
m = mask3d[:,:,S.slice["axial"]].astype(np.uint8)
m = _rot90_if_needed(m)
return m
def _axial_overlay_fill(mask3d, rgba):
sl = _rot90_if_needed(S.vol[:,:,S.slice["axial"]])
base=Image.fromarray((sl*255).astype(np.uint8)).resize((S.render_px,S.render_px),Image.BILINEAR).convert("RGBA")
m2d = _axial_mask2d_rot(mask3d)
if m2d is None: return base.convert("RGB")
m2d = _resize_slice_nearest((m2d>0).astype(np.uint8), S.render_px, S.render_px)
over=np.zeros((S.render_px,S.render_px,4),dtype=np.uint8); over[m2d>0]=rgba
return Image.alpha_composite(base,Image.fromarray(over,"RGBA")).convert("RGB")
def _interaction_2d():
view = S.active_view
if not S.loaded: return None
sl=_slice2d(view)
im=Image.fromarray((sl*255).astype(np.uint8)).resize((S.render_px,S.render_px),Image.BILINEAR).convert("RGB")
w=h=S.render_px
dr=ImageDraw.Draw(im)
_draw_cross(dr, view, w, h)
tol=0
for i, (x,y,z) in enumerate(S.seeds):
on_plane = (
(view=="axial" and abs(z - S.slice["axial"]) <= tol) or
(view=="sagittal" and abs(x - S.slice["sagittal"])<= tol) or
(view=="coronal" and abs(y - S.slice["coronal"]) <= tol)
)
if not on_plane: continue
u,v=_cross_pix_on_rot(view,w,h,x,y,z)
r=4
dr.ellipse((u-r,v-r,u+r,v+r), fill=SEED_RGB, outline=(40,140,100), width=1)
dr.text((u+6, v-8), f"{i+1}", fill=(30,30,30))
if S.bbox_mode:
for i, (x,y,z) in enumerate(S.bbox_points):
on_plane = (
(view=="axial" and abs(z - S.slice["axial"]) <= tol) or
(view=="sagittal" and abs(x - S.slice["sagittal"])<= tol) or
(view=="coronal" and abs(y - S.slice["coronal"]) <= tol)
)
if not on_plane: continue
u,v=_cross_pix_on_rot(view,w,h,x,y,z)
r=6
dr.rectangle((u-r,v-r,u+r,v+r), outline=BBOX_RGB, width=3)
text = "P1" if i == 0 else "P2"
dr.text((u+10, v-10), text, fill=BBOX_RGB)
for (x1,y1,z1,x2,y2,z2) in S.bboxes:
if view=="axial":
curr_z = S.slice["axial"]
if min(z1,z2)-1 <= curr_z <= max(z1,z2)+1:
u1,v1 = _cross_pix_on_rot(view,w,h,x1,y1,curr_z)
u2,v2 = _cross_pix_on_rot(view,w,h,x2,y2,curr_z)
dr.rectangle([u1,v1,u2,v2], outline=(0,255,0), width=2)
return im
# ===================== segmentation ========================================
def _segment_with_model():
if PREDICTOR is None and not _init_predictor_once():
return None, "model-init-failed"
try:
img = S.vol[None].astype(np.float32)
PREDICTOR.reset_interactions()
PREDICTOR.set_image(img)
PREDICTOR.set_target_buffer(np.zeros_like(img[0], np.float32))
PREDICTOR._finish_preprocessing_and_initialize_interactions()
for (x,y,z) in S.seeds:
PREDICTOR.add_point_interaction(x, y, z, foreground=True)
for (x1,y1,z1,x2,y2,z2) in S.bboxes:
PREDICTOR.add_bbox_interaction(
min(x1,x2), min(y1,y2), min(z1,z2),
max(x1,x2), max(y1,y2), max(z1,z2)
)
PREDICTOR._predict_without_interaction()
pred = (PREDICTOR.target_buffer.astype(np.float32) > 0.5).astype(np.uint8)
if pred.shape != S.shape:
print(f"[MODEL] resize pred {pred.shape}->{S.shape}")
pred = _resample_mask_to_vol_shape(pred, S.shape)
return pred, "ok"
except Exception as e:
print(f"[MODEL] inference failed: {e}")
return None, "model-error"
def _segment_fallback():
if (not S.seeds and not S.bboxes) or not HAVE_SCIPY:
return None, "no-interactions-or-scipy"
X,Y,Z=S.shape
field=np.zeros((X,Y,Z),dtype=np.float32)
for (x,y,z) in S.seeds: field[x,y,z]=1.0
for (x1,y1,z1,x2,y2,z2) in S.bboxes:
field[min(x1,x2):max(x1,x2)+1, min(y1,y2):max(y1,y2)+1, min(z1,z2):max(z1,z2)+1] = 0.5
t0=time.time()
prob=ndi.gaussian_filter(field,sigma=6.0)
if prob.max()>0: prob/=prob.max()
nz=prob[prob>0]; thr=np.percentile(nz,70) if nz.size else 0.5
mask=prob>=max(thr,1e-3)
lab,nlab=ndi.label(mask.astype(np.uint8))
if nlab>1:
keep=np.zeros(nlab+1,np.uint8)
for (x,y,z) in S.seeds: keep[lab[x,y,z]]=1
for (x1,y1,z1,x2,y2,z2) in S.bboxes:
xm,ym,zm=(x1+x2)//2,(y1+y2)//2,(z1+z2)//2
keep[lab[xm,ym,zm]]=1
mask=keep[lab]>0
mask=ndi.binary_closing(mask,iterations=1); mask=ndi.binary_opening(mask,iterations=1)
print(f"[FB] seg {time.time()-t0:.3f}s | vox={int(mask.sum())}")
return (mask>0).astype(np.uint8), "ok"
def do_segment():
pred, tag = _segment_with_model()
if pred is None:
pred, tag2 = _segment_fallback(); tag = f"{tag}->{tag2}"
S.pred = pred if pred is not None else None
print(f"[SEG] done: {tag}")
return "OK" if S.pred is not None else "Failed"
def save_prediction():
if S.pred is None:
return "No prediction to save", None
if not HAVE_NIB:
return "nibabel not installed", None
try:
tmp_dir = Path(tempfile.mkdtemp(prefix="menrt_output_"))
out_path = tmp_dir / f"{S.case_id}_pred.nii.gz"
affine = S.ref_affine if S.ref_affine is not None else np.eye(4)
header = S.ref_header.copy() if S.ref_header is not None else None
nii_img = nib.Nifti1Image(S.pred.astype(np.uint8), affine, header=header)
nib.save(nii_img, str(out_path))
print(f"[SAVE] {out_path}")
return "Saved successfully!", str(out_path)
except Exception as e:
print(f"[SAVE] error: {e}")
return f"Save failed: {e}", None
# ---------- helpers ----------
def _seed_rows():
rows=[]
for i,(x,y,z) in enumerate(S.seeds):
v = S.seed_views[i] if i < len(S.seed_views) else ""
rows.append([i+1, "point", v, x, y, z])
for i,(x1,y1,z1,x2,y2,z2) in enumerate(S.bboxes):
rows.append([len(S.seeds)+i+1, "bbox", "3D", f"{x1}-{x2}", f"{y1}-{y2}", f"{z1}-{z2}"])
return rows
def _seed_dropdown_options():
opts=[]
for i,(x,y,z) in enumerate(S.seeds):
v = S.seed_views[i] if i < len(S.seed_views) else "axial"
opts.append(f"{v} → point {i+1} → ({x},{y},{z})")
for i,(x1,y1,z1,x2,y2,z2) in enumerate(S.bboxes):
opts.append(f"3D → bbox {i+1} → ({x1},{y1},{z1})-({x2},{y2},{z2})")
return opts
def _debug_widgets(current_idx=None):
rows = _seed_rows()
df_upd = gr.update(value=rows)
opts = _seed_dropdown_options()
if current_idx is None:
val = (opts[-1] if opts else None)
else:
val = (opts[current_idx] if (0 <= current_idx < len(opts)) else (opts[-1] if opts else None))
dd_upd = gr.update(choices=opts, value=val)
return df_upd, dd_upd
def _figs_and_imgs():
top_ax=render_top("axial")
top_sg=render_top("sagittal")
top_co=render_top("coronal")
ax_gt = _axial_overlay_fill(S.gt, GT_RGBA_FILL)
ax_pr = _axial_overlay_fill(S.pred, PR_RGBA_FILL)
inter2d = _interaction_2d()
return top_ax, top_sg, top_co, ax_gt, ax_pr, inter2d
def _bar_ranges_and_values():
X,Y,Z=S.shape
return (gr.update(minimum=0,maximum=Z-1,value=S.slice["axial"],visible=True),
gr.update(minimum=0,maximum=X-1,value=S.slice["sagittal"],visible=True),
gr.update(minimum=0,maximum=Y-1,value=S.slice["coronal"],visible=True))
def _parse_evt_xy(evt):
"""강화된 이벤트 파싱"""
print(f"[DEBUG_EVT] Event received: {evt}")
print(f"[DEBUG_EVT] Event type: {type(evt)}")
print(f"[DEBUG_EVT] Event dir: {dir(evt)}")
if evt is None:
print("[DEBUG_EVT] Event is None!")
return None
try:
# Method 1: evt.index
if hasattr(evt, "index") and evt.index is not None:
ix = evt.index
print(f"[DEBUG_EVT] Found index: {ix}")
if isinstance(ix, (list, tuple)) and len(ix) >= 2:
result = int(ix[0]), int(ix[1])
print(f"[DEBUG_EVT] Parsed from index: {result}")
return result
# Method 2: evt.x, evt.y
if hasattr(evt, "x") and hasattr(evt, "y"):
x_val = getattr(evt, "x")
y_val = getattr(evt, "y")
print(f"[DEBUG_EVT] Found x={x_val}, y={y_val}")
if x_val is not None and y_val is not None:
result = int(x_val), int(y_val)
print(f"[DEBUG_EVT] Parsed from x,y: {result}")
return result
except Exception as e:
print(f"[DEBUG_EVT] Parse error: {e}")
print("[DEBUG_EVT] Failed to parse coordinates!")
return None
def _disp_to_vol(view,u,v):
X,Y,Z=S.shape; w,h=S.disp_wh[view]
if w<=0 or h<=0: w=h=S.render_px
if view=="axial":
x = int(round(u * (X-1) / max(w-1,1)))
y = int(round((Y-1) - v * (Y-1) / max(h-1,1)))
z = S.slice["axial"]
elif view=="sagittal":
z = int(round(u * (Z-1) / max(w-1,1)))
y = int(round((Y-1) - v * (Y-1) / max(h-1,1)))
x = S.slice["sagittal"]
else:
z = int(round(u * (Z-1) / max(w-1,1)))
x = int(round((X-1) - v * (X-1) / max(h-1,1)))
y = S.slice["coronal"]
x=max(0,min(X-1,x)); y=max(0,min(Y-1,y)); z=max(0,min(Z-1,z))
return x,y,z
def _thumb_from_case(case_id,px=96):
case_dir=DATA_ROOT/case_id
pngs = sorted(glob.glob(str(case_dir / "png_axial" / "*.png")))
if not pngs:
pngs = sorted(glob.glob(str(case_dir / "png_axial" / "*.jpg")))
if pngs:
im = Image.open(pngs[len(pngs)//2]).convert("L")
arr = _norm01(np.array(im).astype(np.float32))
im = Image.fromarray((arr*255).astype(np.uint8)).resize((px,px),Image.BILINEAR)
else:
vol, _, _ = _load_png_stack(case_dir)
if vol is None:
vol, _, _ = _load_nifti(case_dir,case_id,ds=2)
if vol is None:
im = Image.new("L",(px,px),30)
else:
mid = vol[:,:,vol.shape[2]//2]
im = Image.fromarray((mid*255).astype(np.uint8)).resize((px,px),Image.BILINEAR)
im = _rot90_if_needed(im)
return ImageOps.expand(im,border=1,fill=200)
# ---------- callbacks ----------
def on_load(case_id):
load_case(case_id)
preload_model_in_background()
imgs=_figs_and_imgs(); bars=_bar_ranges_and_values(); dbg=_debug_widgets()
return (*imgs,*bars,*dbg, gr.update(value="Point"), None, "Loaded")
def on_gallery_select(evt: gr.SelectData):
idx=0
if hasattr(evt,"index"):
ix=evt.index; idx=int(ix[0] if isinstance(ix,(list,tuple)) else ix)
idx=max(0,min(len(EXAMPLES)-1,idx))
cid=EXAMPLES[idx]
out=on_load(cid)
return (*out, gr.update(value=cid))
def _click_common(view, evt: gr.SelectData):
print(f"[CLICK_HANDLER] Called for view={view}")
print(f"[CLICK_HANDLER] S.loaded={S.loaded}")
print(f"[CLICK_HANDLER] S.bbox_mode={S.bbox_mode}")
if not S.loaded:
print("[CLICK_HANDLER] Not loaded yet!")
return (*_figs_and_imgs(),*_bar_ranges_and_values(),*_debug_widgets(),
gr.update(value="Point" if not S.bbox_mode else "BBox"), None, "Not loaded")
xy=_parse_evt_xy(evt)
if xy is None:
print("[CLICK_HANDLER] Failed to parse event!")
return (*_figs_and_imgs(),*_bar_ranges_and_values(),*_debug_widgets(),
gr.update(value="Point" if not S.bbox_mode else "BBox"), None, "Parse failed")
u,v=xy
x,y,z=_disp_to_vol(view,u,v)
print(f"[CLICK_HANDLER] Clicked at display ({u},{v}) -> volume ({x},{y},{z})")
if S.bbox_mode:
S.bbox_points.append((x,y,z))
if len(S.bbox_points) == 2:
p1, p2 = S.bbox_points
S.bboxes.append((*p1, *p2))
S.bbox_points = []
print(f"[UI] bbox created: {S.bboxes[-1]}")
status = "BBox created!"
else:
print(f"[UI] bbox point {len(S.bbox_points)}/2")
status = f"BBox point {len(S.bbox_points)}/2"
else:
S.seeds.append((x,y,z)); S.seed_views.append(view)
print(f"[UI] seed+ {(x,y,z)} total={len(S.seeds)}")
status = f"Added point {len(S.seeds)}"
S.cross={"x":x,"y":y,"z":z}
S.slice={"sagittal":x,"coronal":y,"axial":z}
S.active_view=view
imgs=_figs_and_imgs(); bars=_bar_ranges_and_values()
idx = len(S.seeds) + len(S.bboxes) - 1
dbg=_debug_widgets(current_idx=idx)
return (*imgs,*bars,*dbg, gr.update(value="Point" if not S.bbox_mode else "BBox"), None, status)
def on_axial_select(evt: gr.SelectData):
print("[EVENT] on_axial_select triggered!")
return _click_common("axial", evt)
def on_sagittal_select(evt: gr.SelectData):
print("[EVENT] on_sagittal_select triggered!")
return _click_common("sagittal", evt)
def on_coronal_select(evt: gr.SelectData):
print("[EVENT] on_coronal_select triggered!")
return _click_common("coronal", evt)
def on_seg_button():
msg=do_segment()
return (*_figs_and_imgs(),*_bar_ranges_and_values(),*_debug_widgets(),
gr.update(value="Point" if not S.bbox_mode else "BBox"), None, f"Segment: {msg}")
def on_clear():
S.seeds=[]; S.seed_views=[]; S.pred=None
S.bboxes=[]; S.bbox_points=[]
print("[UI] cleared all")
return (*_figs_and_imgs(),*_bar_ranges_and_values(),*_debug_widgets(),
gr.update(value="Point"), None, "Cleared")
def on_undo():
if S.bbox_mode and S.bbox_points:
S.bbox_points.pop()
elif S.bboxes:
S.bboxes.pop()
elif S.seeds:
S.seeds.pop(); S.seed_views.pop() if S.seed_views else None
print("[UI] undo")
return (*_figs_and_imgs(),*_bar_ranges_and_values(),*_debug_widgets(),
gr.update(value="Point" if not S.bbox_mode else "BBox"), None, "Undo")
def on_mode_toggle(mode):
S.bbox_mode = (mode == "BBox")
S.bbox_points = []
print(f"[UI] mode -> {mode}")
return (*_figs_and_imgs(),*_bar_ranges_and_values(),*_debug_widgets(),
gr.update(value=mode), None, f"Mode: {mode}")
def on_save():
msg, path = save_prediction()
if path:
return (*_figs_and_imgs(),*_bar_ranges_and_values(),*_debug_widgets(),
gr.update(value="Point" if not S.bbox_mode else "BBox"),
gr.update(visible=True, value=path), msg)
return (*_figs_and_imgs(),*_bar_ranges_and_values(),*_debug_widgets(),
gr.update(value="Point" if not S.bbox_mode else "BBox"),
gr.update(visible=False, value=None), msg)
def on_z_release(z_idx):
S.slice["axial"]=int(z_idx); S.cross["z"]=S.slice["axial"]; S.active_view="axial"
return (*_figs_and_imgs(),*_bar_ranges_and_values(),*_debug_widgets(),
gr.update(value="Point" if not S.bbox_mode else "BBox"), None, f"Z={z_idx}")
def on_x_release(x_idx):
S.slice["sagittal"]=int(x_idx); S.cross["x"]=S.slice["sagittal"]; S.active_view="sagittal"
return (*_figs_and_imgs(),*_bar_ranges_and_values(),*_debug_widgets(),
gr.update(value="Point" if not S.bbox_mode else "BBox"), None, f"X={x_idx}")
def on_y_release(y_idx):
S.slice["coronal"]=int(y_idx); S.cross["y"]=S.slice["coronal"]; S.active_view="coronal"
return (*_figs_and_imgs(),*_bar_ranges_and_values(),*_debug_widgets(),
gr.update(value="Point" if not S.bbox_mode else "BBox"), None, f"Y={y_idx}")
def _jump_to_idx(idx:int):
total = len(S.seeds) + len(S.bboxes)
if not (0 <= idx < total):
return (*_figs_and_imgs(),*_bar_ranges_and_values(),*_debug_widgets(),
gr.update(value="Point" if not S.bbox_mode else "BBox"), None, "")
if idx < len(S.seeds):
x,y,z = S.seeds[idx]
view = S.seed_views[idx] if idx < len(S.seed_views) else "axial"
S.cross={"x":x,"y":y,"z":z}
S.slice={"sagittal":x,"coronal":y,"axial":z}
S.active_view=view
return (*_figs_and_imgs(),*_bar_ranges_and_values(),*_debug_widgets(current_idx=idx),
gr.update(value="Point" if not S.bbox_mode else "BBox"), None, f"Jump to {idx+1}")
def on_seed_df_select(evt):
try:
row = int(evt.index[0]) if hasattr(evt, "index") else 0
except Exception:
row = 0
return _jump_to_idx(row)
def on_seed_dd_change(val):
if not val:
return (*_figs_and_imgs(),*_bar_ranges_and_values(),*_debug_widgets(),
gr.update(value="Point" if not S.bbox_mode else "BBox"), None, "")
try:
if "point" in val:
p = val.split("point")[1].strip()
num = int(p.split("→")[0].strip())
idx = num - 1
else:
idx = len(S.seeds) + len(S.bboxes) - 1
except Exception:
idx = len(S.seeds) + len(S.bboxes) - 1
return _jump_to_idx(idx)
# ---------- UI ----------
css = """
:root {
--bg:#f7fbff; --card:#ffffff; --accent:#1e90ff; --shadow:0 8px 26px rgba(45,156,219,.12);
}
.gradio-container{font-family:Inter,ui-sans-serif,system-ui;background:var(--bg)}
.round{background:var(--card);border-radius:16px;padding:10px;box-shadow:var(--shadow);border:1px solid #e8f0fb}
.section{font-weight:700;color:#114b8b;margin:2px 0 8px}
.tiny .gr-slider input[type="range"]{height:6px}
.tiny .gr-form{gap:6px}
.smallnote{color:#3e6285;font-size:12px;margin-top:6px}
"""
THUMBS=[_thumb_from_case(cid,px=96) for cid in EXAMPLES]
with gr.Blocks(css=css, title="Interactive-MEN-RT", theme=gr.themes.Soft(), analytics_enabled=False) as demo:
gr.Markdown("""
<h3 style='color:#114b8b'>Interactive-MEN-RT Segmentation</h3>
<div class='smallnote'>
Domain-Specialized Interactive Segmentation for Meningioma Radiotherapy Planning<br>
<b>Research only — Not for clinical use</b>
</div>
""")
with gr.Row():
with gr.Column(scale=1, min_width=280, elem_classes=["round"]):
gr.Markdown("<div class='section'>Demo Case</div>")
gallery = gr.Gallery(value=THUMBS, columns=1, height=110, allow_preview=False, preview=False, show_label=False)
case_dd = gr.Dropdown(choices=EXAMPLES, value=EXAMPLES[0], label="Case")
mode_radio = gr.Radio(["Point", "BBox"], value="Point", label="Interaction Mode")
with gr.Row():
seg_btn = gr.Button("Segment", variant="primary")
save_btn = gr.Button("Save NIfTI", variant="secondary")
with gr.Row():
undo_btn = gr.Button("Undo")
clr_btn = gr.Button("Clear")
gr.Markdown("<div class='section' style='margin-top:8px'>Interactions</div>")
seeds_df = gr.Dataframe(
headers=["#", "type", "view", "x", "y", "z"],
value=[],
datatype=["number","str","str","str","str","str"],
interactive=False,
wrap=True,
row_count=(0, "dynamic")
)
seed_dd = gr.Dropdown(choices=[], value=None, label="Go to point")
pred_file = gr.File(label="Download Prediction", visible=False)
status_text = gr.Textbox(label="Status", value="", interactive=False, lines=1)
with gr.Column(scale=5):
with gr.Row():
with gr.Column(elem_classes=["round"]):
axial = gr.Image(type="pil", interactive=True, height=RENDER_PX_DEFAULT+8, label="Axial (Z)")
z_bar = gr.Slider(0,1,value=0,step=1,label="Z", elem_classes=["tiny"])
with gr.Column(elem_classes=["round"]):
sagittal = gr.Image(type="pil", interactive=True, height=RENDER_PX_DEFAULT+8, label="Sagittal (X)")
x_bar = gr.Slider(0,1,value=0,step=1,label="X", elem_classes=["tiny"])
with gr.Column(elem_classes=["round"]):
coronal = gr.Image(type="pil", interactive=True, height=RENDER_PX_DEFAULT+8, label="Coronal (Y)")
y_bar = gr.Slider(0,1,value=0,step=1,label="Y", elem_classes=["tiny"])
with gr.Row():
out_ax_gt = gr.Image(type="pil", interactive=False, height=RENDER_PX_DEFAULT+8, label="Ground Truth")
out_ax_pr = gr.Image(type="pil", interactive=False, height=RENDER_PX_DEFAULT+8, label="Prediction")
inter2d = gr.Image(type="pil", interactive=False, height=RENDER_PX_DEFAULT+8, label="Interactions")
outputs = [axial, sagittal, coronal, out_ax_gt, out_ax_pr, inter2d,
z_bar, x_bar, y_bar, seeds_df, seed_dd, mode_radio, pred_file, status_text]
demo.load(lambda: on_load(EXAMPLES[0]), [], outputs)
case_dd.change(lambda cid: on_load(cid), [case_dd], outputs)
gallery.select(on_gallery_select, [], outputs + [case_dd])
# 클릭 이벤트 - 강화된 핸들러
axial.select(on_axial_select, [], outputs)
sagittal.select(on_sagittal_select, [], outputs)
coronal.select(on_coronal_select, [], outputs)
z_bar.release(on_z_release, [z_bar], outputs)
x_bar.release(on_x_release, [x_bar], outputs)
y_bar.release(on_y_release, [y_bar], outputs)
mode_radio.change(on_mode_toggle, [mode_radio], outputs)
seg_btn.click(on_seg_button, [], outputs)
save_btn.click(on_save, [], outputs)
clr_btn.click(on_clear, [], outputs)
undo_btn.click(on_undo, [], outputs)
seeds_df.select(on_seed_df_select, [], outputs)
seed_dd.change(on_seed_dd_change, [seed_dd], outputs)
if __name__ == "__main__":
demo.launch(server_name="0.0.0.0", server_port=7860, show_error=True, share=True,
allowed_paths=[str(DATA_ROOT)]) |