File size: 22,456 Bytes
cacaeb4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b95781b
cacaeb4
 
 
 
 
 
 
 
 
 
 
 
 
7ea83d4
cacaeb4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7ea83d4
 
 
 
 
 
 
 
 
 
 
 
169758c
 
 
7ea83d4
 
 
 
 
 
 
 
 
 
 
169758c
 
 
 
 
 
 
7ea83d4
169758c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7ea83d4
 
 
 
cacaeb4
 
 
 
 
 
 
 
 
 
 
 
7ea83d4
 
cacaeb4
 
 
 
 
 
b95781b
cacaeb4
 
 
b95781b
cacaeb4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0ea4706
 
 
 
 
 
 
cacaeb4
 
 
 
 
 
 
 
 
 
 
0ea4706
 
 
 
 
cacaeb4
 
 
 
 
 
 
 
0ea4706
cacaeb4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b95781b
 
 
 
 
 
 
 
 
 
 
 
 
 
0ea4706
 
 
cacaeb4
0ea4706
cacaeb4
 
 
 
 
0ea4706
 
 
 
 
 
 
 
cacaeb4
0ea4706
f50f28a
 
cacaeb4
 
 
0ea4706
 
 
cacaeb4
 
 
0ea4706
 
 
cacaeb4
 
 
 
 
 
0ea4706
 
 
cacaeb4
 
 
 
 
 
0ea4706
 
cacaeb4
 
0ea4706
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cacaeb4
 
 
b95781b
 
0ea4706
f50f28a
 
 
 
 
 
 
 
 
 
cacaeb4
 
 
 
 
 
 
 
 
 
 
 
 
0ea4706
 
 
 
 
 
 
 
 
 
cacaeb4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
169758c
cacaeb4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b95781b
cacaeb4
b95781b
cacaeb4
 
b95781b
cacaeb4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b95781b
cacaeb4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
169758c
cacaeb4
 
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
#!/usr/bin/env python3
"""
Fall Detection Gradio App

YOLOv11-Pose + ST-GCN 2-stage νŒŒμ΄ν”„λΌμΈμ„ μ‚¬μš©ν•œ 낙상 감지 데λͺ¨μž…λ‹ˆλ‹€.
HF Spaces Zero GPU ν™˜κ²½μ—μ„œ μ‹€ν–‰λ©λ‹ˆλ‹€.

μ‚¬μš©λ²• (둜컬):
    python demo_gradio/app.py

μ‚¬μš©λ²• (HF Spaces):
    μžλ™μœΌλ‘œ app.pyκ°€ μ‹€ν–‰λ©λ‹ˆλ‹€.

μž‘μ„±μž: Fall Detection Pipeline Team
μž‘μ„±μΌ: 2025-11-26
"""

import os
import subprocess
import sys
import tempfile
import time
from pathlib import Path
from typing import Iterable, Optional, Tuple

import cv2
import gradio as gr
import numpy as np
import plotly.graph_objects as go
import torch
from gradio.themes import Soft
from gradio.themes.utils import colors, fonts, sizes
from huggingface_hub import hf_hub_download

# ν”„λ‘œμ νŠΈ 루트λ₯Ό Python path에 μΆ”κ°€
# pipeline/demo_gradio/app.py -> pipeline -> project_root
PROJECT_ROOT = Path(__file__).parent.parent.parent
sys.path.insert(0, str(PROJECT_ROOT))

# Zero GPU ν˜Έν™˜ μ„€μ •
try:
    import spaces
    SPACES_AVAILABLE = True
except ImportError:
    SPACES_AVAILABLE = False


# -----------------------------------------------------------------------------
# μ»€μŠ€ν…€ ν…Œλ§ˆ (PRITHIVSAKTHIUR μŠ€νƒ€μΌ)
# -----------------------------------------------------------------------------
colors.custom_color = colors.Color(
    name="custom_color",
    c50="#EBF3F8", c100="#D3E5F0", c200="#A8CCE1",
    c300="#7DB3D2", c400="#529AC3", c500="#4682B4",
    c600="#3E72A0", c700="#36638C", c800="#2E5378",
    c900="#264364", c950="#1E3450",
)


class CustomTheme(Soft):
    def __init__(
        self,
        *,
        primary_hue: colors.Color | str = colors.gray,
        secondary_hue: colors.Color | str = colors.custom_color,
        neutral_hue: colors.Color | str = colors.slate,
        text_size: sizes.Size | str = sizes.text_lg,
        font: fonts.Font | str | Iterable[fonts.Font | str] = (
            fonts.GoogleFont("Outfit"), "Arial", "sans-serif",
        ),
        font_mono: fonts.Font | str | Iterable[fonts.Font | str] = (
            fonts.GoogleFont("IBM Plex Mono"), "ui-monospace", "monospace",
        ),
    ):
        super().__init__(
            primary_hue=primary_hue,
            secondary_hue=secondary_hue,
            neutral_hue=neutral_hue,
            text_size=text_size,
            font=font,
            font_mono=font_mono,
        )
        super().set(
            background_fill_primary="*primary_50",
            body_background_fill="linear-gradient(135deg, *primary_200, *primary_100)",
            button_primary_text_color="white",
            button_primary_background_fill="linear-gradient(90deg, *secondary_500, *secondary_600)",
            button_primary_background_fill_hover="linear-gradient(90deg, *secondary_600, *secondary_700)",
            slider_color="*secondary_500",
            block_title_text_weight="600",
            block_border_width="3px",
            block_shadow="*shadow_drop_lg",
            button_primary_shadow="*shadow_drop_lg",
        )


custom_theme = CustomTheme()

# -----------------------------------------------------------------------------
# CSS μŠ€νƒ€μΌ
# -----------------------------------------------------------------------------
css = """
#col-container { margin: 0 auto; max-width: 1200px; }
#main-title h1 { font-size: 2.3em !important; }
.submit-btn {
    background-color: #4682B4 !important;
    color: white !important;
}
.submit-btn:hover {
    background-color: #5A9BD4 !important;
}
.result-label {
    font-size: 1.5em !important;
    font-weight: bold !important;
    padding: 10px !important;
    border-radius: 8px !important;
}
.fall-detected {
    background-color: #FF4444 !important;
    color: white !important;
}
.non-fall {
    background-color: #44BB44 !important;
    color: white !important;
}
"""

# -----------------------------------------------------------------------------
# λ””λ°”μ΄μŠ€ μ„€μ •
# -----------------------------------------------------------------------------
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")


# -----------------------------------------------------------------------------
# GPU λ°μ½”λ ˆμ΄ν„° (둜컬/HF Spaces ν˜Έν™˜)
# -----------------------------------------------------------------------------
def gpu_decorator(duration: int = 120):
    """λ‘œμ»¬μ—μ„œλŠ” κ·Έλƒ₯ μ‹€ν–‰, Spacesμ—μ„œλŠ” GPU ν• λ‹Ή"""
    def decorator(func):
        if SPACES_AVAILABLE:
            return spaces.GPU(duration=duration)(func)
        return func
    return decorator


# -----------------------------------------------------------------------------
# λͺ¨λΈ λ‹€μš΄λ‘œλ“œ (HuggingFace Hub)
# -----------------------------------------------------------------------------
HF_MODEL_REPO = "YoungjaeDev/fall-detection-models"


def download_models() -> tuple[str, str]:
    """
    HuggingFace Hubμ—μ„œ λͺ¨λΈ λ‹€μš΄λ‘œλ“œ (μΊμ‹œλ¨)

    Returns:
        tuple: (pose_model_path, stgcn_checkpoint_path)

    Raises:
        RuntimeError: λͺ¨λΈ λ‹€μš΄λ‘œλ“œ λ˜λŠ” 검증 μ‹€νŒ¨ μ‹œ
    """
    # 둜컬 경둜 μš°μ„  확인 (개발 ν™˜κ²½)
    local_pose = Path("yolo11m-pose.pt")
    local_stgcn = Path("runs/stgcn_binary_exp2_fixed_graph/best_acc.pth")

    if local_pose.exists() and local_stgcn.exists():
        return str(local_pose), str(local_stgcn)

    # HuggingFace Hubμ—μ„œ λ‹€μš΄λ‘œλ“œ (Private repoλŠ” HF_TOKEN ν™˜κ²½λ³€μˆ˜ ν•„μš”)
    token = os.environ.get("HF_TOKEN")

    # Private μ €μž₯μ†Œ 접근을 μœ„ν•œ 토큰 확인
    if token is None:
        raise RuntimeError(
            "HF_TOKEN ν™˜κ²½λ³€μˆ˜κ°€ μ„€μ •λ˜μ§€ μ•Šμ•˜μŠ΅λ‹ˆλ‹€. "
            "Private λͺ¨λΈ μ €μž₯μ†Œ 접근을 μœ„ν•΄ HF_TOKEN이 ν•„μš”ν•©λ‹ˆλ‹€. "
            "HF Spaces의 경우 Settings > Secretsμ—μ„œ μ„€μ •ν•˜μ„Έμš”."
        )

    try:
        pose_model_path = hf_hub_download(
            repo_id=HF_MODEL_REPO,
            filename="yolo11m-pose.pt",
            token=token
        )

        stgcn_checkpoint = hf_hub_download(
            repo_id=HF_MODEL_REPO,
            filename="best_acc.pth",
            token=token
        )
    except Exception as e:
        raise RuntimeError(
            f"λͺ¨λΈ λ‹€μš΄λ‘œλ“œ μ‹€νŒ¨: {e}\n"
            f"μ €μž₯μ†Œ: {HF_MODEL_REPO}\n"
            f"HF_TOKEN이 μ˜¬λ°”λ₯΄κ²Œ μ„€μ •λ˜μ—ˆλŠ”μ§€ ν™•μΈν•˜μ„Έμš”."
        ) from e

    # λ‹€μš΄λ‘œλ“œλœ 파일 검증
    pose_path = Path(pose_model_path)
    stgcn_path = Path(stgcn_checkpoint)

    if not pose_path.exists():
        raise RuntimeError(f"Pose λͺ¨λΈ 파일이 μ‘΄μž¬ν•˜μ§€ μ•ŠμŠ΅λ‹ˆλ‹€: {pose_model_path}")
    if not stgcn_path.exists():
        raise RuntimeError(f"ST-GCN 체크포인트 파일이 μ‘΄μž¬ν•˜μ§€ μ•ŠμŠ΅λ‹ˆλ‹€: {stgcn_checkpoint}")

    # 파일 크기 검증 (λ„ˆλ¬΄ μž‘μœΌλ©΄ μ†μƒλœ 파일일 κ°€λŠ₯μ„±)
    pose_size = pose_path.stat().st_size
    stgcn_size = stgcn_path.stat().st_size
    if pose_size < 1_000_000:  # 1MB 미만
        raise RuntimeError(f"Pose λͺ¨λΈ 파일이 λ„ˆλ¬΄ μž‘μŠ΅λ‹ˆλ‹€: {pose_size} bytes")
    if stgcn_size < 1_000_000:  # 1MB 미만
        raise RuntimeError(f"ST-GCN 체크포인트 파일이 λ„ˆλ¬΄ μž‘μŠ΅λ‹ˆλ‹€: {stgcn_size} bytes")

    return pose_model_path, stgcn_checkpoint


# -----------------------------------------------------------------------------
# νŒŒμ΄ν”„λΌμΈ μ΄ˆκΈ°ν™” (μ§€μ—° λ‘œλ”©)
# -----------------------------------------------------------------------------
_pipeline = None


def get_pipeline():
    """νŒŒμ΄ν”„λΌμΈ 싱글톀 λ°˜ν™˜ (μ§€μ—° λ‘œλ”©)"""
    global _pipeline
    if _pipeline is None:
        from pipeline.core.pipeline import FallDetectionPipeline

        # λͺ¨λΈ λ‹€μš΄λ‘œλ“œ (μΊμ‹œλ¨)
        pose_model_path, stgcn_checkpoint = download_models()

        _pipeline = FallDetectionPipeline(
            pose_model_path=pose_model_path,
            stgcn_checkpoint=stgcn_checkpoint,
            window_size=60,
            conf_threshold=0.5,
            fall_threshold=0.85,  # κ°€μ΄λ“œλΌμΈ ꢌμž₯: 0.8-0.9 (false positive <5%)
            temporal_window=5,
            stgcn_stride=5,
            alert_duration=150,
            post_fall_frames=15,  # 2.5초 @ 30fps with stride=5 (κ°€μ΄λ“œλΌμΈ: 2-3초)
            device=str(device),
            debug=False,
            headless=False,
            viz_keypoints="all",
            viz_scale=1.0,
            viz_optimized=True
        )
    return _pipeline


# -----------------------------------------------------------------------------
# ν™•λ₯  κ·Έλž˜ν”„ 생성
# -----------------------------------------------------------------------------
def create_probability_graph(
    frame_indices: list,
    probabilities: list,
    fall_threshold: float = 0.7
) -> go.Figure:
    """
    낙상 ν™•λ₯  κ·Έλž˜ν”„ 생성

    Args:
        frame_indices: ν”„λ ˆμž„ 인덱슀 리슀트
        probabilities: 낙상 ν™•λ₯  리슀트 (0.0-1.0)
        fall_threshold: 낙상 νŒμ • μž„κ³„κ°’

    Returns:
        Plotly Figure 객체
    """
    fig = go.Figure()

    # ν™•λ₯  라인
    fig.add_trace(go.Scatter(
        x=frame_indices,
        y=probabilities,
        mode='lines',
        name='Fall Probability',
        line=dict(color='#4682B4', width=2),
        fill='tozeroy',
        fillcolor='rgba(70, 130, 180, 0.3)'
    ))

    # μž„κ³„κ°’ 라인
    fig.add_hline(
        y=fall_threshold,
        line_dash="dash",
        line_color="red",
        annotation_text=f"Threshold ({fall_threshold})",
        annotation_position="right"
    )

    # λ ˆμ΄μ•„μ›ƒ
    fig.update_layout(
        title="Fall Detection Probability Over Time",
        xaxis_title="Frame",
        yaxis_title="Probability",
        yaxis=dict(range=[0, 1]),
        template="plotly_white",
        height=300,
        margin=dict(l=50, r=50, t=50, b=50),
        showlegend=True,
        legend=dict(
            orientation="h",
            yanchor="bottom",
            y=1.02,
            xanchor="right",
            x=1
        )
    )

    return fig


# -----------------------------------------------------------------------------
# 슀마트 클립 μΆ”μΆœ μ„€μ • (Issue #82)
# -----------------------------------------------------------------------------
CLIP_PRE_FALL_SECONDS = 1.0   # 낙상 μ „ 1초
CLIP_POST_FALL_SECONDS = 2.0  # 낙상 ν›„ 2초


# -----------------------------------------------------------------------------
# 메인 μΆ”λ‘  ν•¨μˆ˜
# -----------------------------------------------------------------------------
@gpu_decorator(duration=120)
def process_video(
    video_path: str,
    fall_threshold: float,
    viz_keypoints: str,
    progress: gr.Progress = gr.Progress()
) -> Tuple[Optional[str], Optional[go.Figure], str]:
    """
    λΉ„λ””μ˜€ 처리 및 낙상 감지 (슀마트 클립 μΆ”μΆœ)

    Issue #82: 낙상 감지 κ΅¬κ°„λ§Œ 클립으둜 μΆ”μΆœν•˜μ—¬ 인코딩 μ‹œκ°„ λŒ€ν­ κ°μ†Œ
    - 낙상 감지 μ‹œ: 낙상 μ „ 1초 + 낙상 ν›„ 2초 κ΅¬κ°„λ§Œ μΆ”μΆœ
    - 비낙상 μ‹œ: 낙상 미감지 λ©”μ‹œμ§€ λ°˜ν™˜

    Args:
        video_path: μž…λ ₯ λΉ„λ””μ˜€ 경둜
        fall_threshold: 낙상 νŒμ • μž„κ³„κ°’ (0.0-1.0)
        viz_keypoints: ν‚€ν¬μΈνŠΈ ν‘œμ‹œ λͺ¨λ“œ ('all' λ˜λŠ” 'major')
        progress: Gradio μ§„ν–‰λ₯  ν‘œμ‹œ

    Returns:
        output_video_path: κ²°κ³Ό 클립 경둜 (낙상 감지 μ‹œ) λ˜λŠ” None (비낙상)
        probability_graph: ν™•λ₯  κ·Έλž˜ν”„
        result_text: μ΅œμ’… νŒμ • ν…μŠ€νŠΈ
    """
    if video_path is None:
        return None, None, "λΉ„λ””μ˜€λ₯Ό μ—…λ‘œλ“œν•΄μ£Όμ„Έμš”."

    try:
        # νŒŒμ΄ν”„λΌμΈ λ‘œλ“œ
        progress(0.1, desc="λͺ¨λΈ λ‘œλ”© 쀑...")
        pipeline = get_pipeline()
        pipeline.fall_threshold = fall_threshold
        pipeline.stgcn_classifier.fall_threshold = fall_threshold
        pipeline.viz_keypoints = viz_keypoints
        pipeline.reset()

        # λΉ„λ””μ˜€ μ—΄κΈ°
        progress(0.2, desc="λΉ„λ””μ˜€ μ—΄κΈ°...")
        cap = cv2.VideoCapture(video_path)
        if not cap.isOpened():
            return None, None, "λΉ„λ””μ˜€λ₯Ό μ—΄ 수 μ—†μŠ΅λ‹ˆλ‹€."

        # λΉ„λ””μ˜€ 정보
        fps = cap.get(cv2.CAP_PROP_FPS)
        width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
        height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
        total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))

        # λΉ„λ””μ˜€ 길이 검증 (120s GPU νƒ€μž„μ•„μ›ƒ λŒ€λΉ„)
        if fps > 0:
            video_duration = total_frames / fps
            # 처리 μ‹œκ°„ μΆ”μ •: λŒ€λž΅ μ‹€μ‹œκ°„μ˜ 1.5λ°° + 인코딩 10초
            estimated_time = video_duration * 1.5 + 10
            if estimated_time > 110:  # 120s νƒ€μž„μ•„μ›ƒμ— μ—¬μœ  두기
                cap.release()
                return None, None, (
                    f"λΉ„λ””μ˜€κ°€ λ„ˆλ¬΄ κΉλ‹ˆλ‹€. "
                    f"λΉ„λ””μ˜€ 길이: {video_duration:.1f}초, "
                    f"μ˜ˆμƒ 처리 μ‹œκ°„: {estimated_time:.1f}초 (μ œν•œ: 110초). "
                    f"60초 μ΄λ‚΄μ˜ λΉ„λ””μ˜€λ₯Ό μ—…λ‘œλ“œν•˜μ„Έμš”."
                )

        # 클립 μΆ”μΆœμ„ μœ„ν•œ ν”„λ ˆμž„ 수 계산
        pre_fall_frames = int(fps * CLIP_PRE_FALL_SECONDS)
        post_fall_frames = int(fps * CLIP_POST_FALL_SECONDS)

        # 처리 루프 - ν”„λ ˆμž„ 버퍼링 + 낙상 감지
        frame_idx = 0
        frame_indices = []
        probabilities = []
        max_confidence = 0.0

        # 낙상 감지 좔적
        first_fall_frame = None  # 첫 낙상 감지 ν”„λ ˆμž„
        fall_detected = False

        # μ‹œκ°ν™” ν”„λ ˆμž„ 버퍼 (클립 μΆ”μΆœμš©)
        vis_frame_buffer = []
        raw_frame_buffer = []  # 원본 ν”„λ ˆμž„ 버퍼 (재처리용)

        while True:
            # ν”„λ ˆμž„ 읽기
            with pipeline.profiler.profile('video_read'):
                ret, frame = cap.read()
            if not ret:
                break

            # 원본 ν”„λ ˆμž„ 버퍼에 μ €μž₯ (클립 μΆ”μΆœμ— ν•„μš”)
            raw_frame_buffer.append(frame.copy())

            # ν”„λ ˆμž„ 처리
            vis_frame, info = pipeline.process_frame(frame, frame_idx)

            # μ‹œκ°ν™” ν”„λ ˆμž„ 버퍼에 μ €μž₯
            vis_frame_buffer.append(vis_frame)

            # ν™•λ₯  기둝
            if info['confidence'] is not None:
                frame_indices.append(frame_idx)
                probabilities.append(info['confidence'])
                max_confidence = max(max_confidence, info['confidence'])

            # 첫 낙상 감지 μ‹œμ  기둝
            if info['alert'] and first_fall_frame is None:
                first_fall_frame = frame_idx
                fall_detected = True

            frame_idx += 1

            # μ§„ν–‰λ₯  μ—…λ°μ΄νŠΈ
            if frame_idx % 10 == 0:
                progress_val = 0.2 + 0.6 * (frame_idx / total_frames)
                progress(progress_val, desc=f"뢄석 쀑... ({frame_idx}/{total_frames})")

        cap.release()

        # ν™•λ₯  κ·Έλž˜ν”„ 생성 (항상 생성)
        progress(0.85, desc="κ·Έλž˜ν”„ 생성 쀑...")
        if frame_indices and probabilities:
            fig = create_probability_graph(frame_indices, probabilities, fall_threshold)
        else:
            fig = None

        # 낙상 미감지 μ‹œ 클립 없이 λ°˜ν™˜
        if not fall_detected or first_fall_frame is None:
            progress(1.0, desc="μ™„λ£Œ!")
            result_text = (
                f"[Non-Fall] 낙상이 κ°μ§€λ˜μ§€ μ•Šμ•˜μŠ΅λ‹ˆλ‹€.\n"
                f"μ΅œλŒ€ ν™•λ₯ : {max_confidence:.1%}\n"
                f"뢄석 ν”„λ ˆμž„: {total_frames}개"
            )
            return None, fig, result_text

        # 클립 ꡬ간 계산
        clip_start = max(0, first_fall_frame - pre_fall_frames)
        clip_end = min(len(vis_frame_buffer), first_fall_frame + post_fall_frames)
        clip_frames = vis_frame_buffer[clip_start:clip_end]

        if not clip_frames:
            progress(1.0, desc="μ™„λ£Œ!")
            return None, fig, "클립 μΆ”μΆœμ— μ‹€νŒ¨ν–ˆμŠ΅λ‹ˆλ‹€."

        # 클립 λΉ„λ””μ˜€ 생성 (ν”„λ ˆμž„ 수 κ°μ†Œλ‘œ 인코딩 μ‹œκ°„ λŒ€ν­ κ°μ†Œ)
        progress(0.9, desc="클립 인코딩 쀑...")
        with tempfile.NamedTemporaryFile(suffix=".mp4", delete=False) as tmp:
            output_path = tmp.name

        fourcc = cv2.VideoWriter_fourcc(*'mp4v')
        # Info panel μΆ”κ°€λ‘œ 높이 80px 증가
        clip_height, clip_width = clip_frames[0].shape[:2]
        out = cv2.VideoWriter(output_path, fourcc, fps, (clip_width, clip_height))

        for vis_frame in clip_frames:
            out.write(vis_frame)
        out.release()

        # H.264 μ½”λ±μœΌλ‘œ μž¬μΈμ½”λ”© (λΈŒλΌμš°μ € ν˜Έν™˜)
        with tempfile.NamedTemporaryFile(suffix=".mp4", delete=False) as tmp:
            output_h264 = tmp.name

        with pipeline.profiler.profile('ffmpeg_encode'):
            subprocess.run(
                [
                    'ffmpeg', '-y', '-i', output_path,
                    '-c:v', 'libx264', '-preset', 'fast', '-crf', '23',
                    output_h264, '-loglevel', 'quiet'
                ],
                check=False,
                capture_output=True
            )

        # mp4v μž„μ‹œ 파일 μ‚­μ œ
        if os.path.exists(output_path):
            os.remove(output_path)

        # H.264 λ³€ν™˜ 성곡 μ—¬λΆ€ 확인
        if os.path.exists(output_h264):
            final_output = output_h264
        else:
            final_output = output_path  # 폴백

        # μ΅œμ’… νŒμ •
        progress(1.0, desc="μ™„λ£Œ!")
        fall_time = first_fall_frame / fps if fps > 0 else 0
        clip_duration = len(clip_frames) / fps if fps > 0 else 0
        result_text = (
            f"[FALL DETECTED] 낙상이 κ°μ§€λ˜μ—ˆμŠ΅λ‹ˆλ‹€!\n"
            f"낙상 μ‹œμ : {fall_time:.2f}초 (ν”„λ ˆμž„ #{first_fall_frame})\n"
            f"μ΅œλŒ€ ν™•λ₯ : {max_confidence:.1%}\n"
            f"클립 길이: {clip_duration:.1f}초 ({len(clip_frames)}ν”„λ ˆμž„)\n"
            f"원본 λŒ€λΉ„: {len(clip_frames)}/{total_frames}ν”„λ ˆμž„ "
            f"({len(clip_frames)/total_frames*100:.1f}% 인코딩)"
        )

        return final_output, fig, result_text

    except Exception as e:
        import traceback
        error_msg = f"처리 쀑 였λ₯˜ λ°œμƒ: {str(e)}\n{traceback.format_exc()}"
        return None, None, error_msg


# -----------------------------------------------------------------------------
# Gradio UI
# -----------------------------------------------------------------------------
def create_demo() -> gr.Blocks:
    """Gradio 데λͺ¨ 생성"""

    with gr.Blocks(theme=custom_theme, css=css) as demo:
        gr.Markdown(
            """
            # Fall Detection Demo

            YOLOv11-Pose + ST-GCN 2-stage νŒŒμ΄ν”„λΌμΈμ„ μ‚¬μš©ν•œ μ‹€μ‹œκ°„ 낙상 감지 데λͺ¨μž…λ‹ˆλ‹€.
            λΉ„λ””μ˜€λ₯Ό μ—…λ‘œλ“œν•˜λ©΄ 낙상 μ—¬λΆ€λ₯Ό λΆ„μ„ν•˜κ³ , κ²°κ³Ό λΉ„λ””μ˜€μ™€ ν™•λ₯  κ·Έλž˜ν”„λ₯Ό μ œκ³΅ν•©λ‹ˆλ‹€.

            **νŒŒμ΄ν”„λΌμΈ ꡬ성:**
            - Stage 1: YOLOv11m-pose (Pose Estimation)
            - Stage 2: ST-GCN (Temporal Classification)
            - Window Size: 60 frames (2초 @ 30fps)
            """,
            elem_id="main-title"
        )

        with gr.Row():
            with gr.Column(scale=1):
                # μž…λ ₯ μ„Ήμ…˜
                gr.Markdown("### μž…λ ₯")
                video_input = gr.Video(
                    label="λΉ„λ””μ˜€ μ—…λ‘œλ“œ",
                    sources=["upload"],
                )

                with gr.Accordion("κ³ κΈ‰ μ„€μ •", open=False):
                    fall_threshold = gr.Slider(
                        minimum=0.7,
                        maximum=0.95,
                        value=0.85,
                        step=0.05,
                        label="낙상 νŒμ • μž„κ³„κ°’",
                        info="ꢌμž₯: 0.8-0.9 (false positive <5% λͺ©ν‘œ)"
                    )
                    viz_keypoints = gr.Radio(
                        choices=["all", "major"],
                        value="all",
                        label="ν‚€ν¬μΈνŠΈ ν‘œμ‹œ",
                        info="all: 전체 17개, major: μ£Όμš” 9개"
                    )

                submit_btn = gr.Button(
                    "뢄석 μ‹œμž‘",
                    variant="primary",
                    elem_classes="submit-btn"
                )

            with gr.Column(scale=1):
                # 좜λ ₯ μ„Ήμ…˜
                gr.Markdown("### κ²°κ³Ό")
                result_text = gr.Textbox(
                    label="νŒμ • κ²°κ³Ό",
                    lines=2,
                    interactive=False
                )
                video_output = gr.Video(
                    label="κ²°κ³Ό λΉ„λ””μ˜€",
                )
                prob_graph = gr.Plot(
                    label="낙상 ν™•λ₯  κ·Έλž˜ν”„",
                )

        # 예제 λΉ„λ””μ˜€
        gr.Markdown("### 예제 λΉ„λ””μ˜€")
        example_dir = Path(__file__).parent / "examples"
        examples = []
        if example_dir.exists():
            for ext in ["*.mp4", "*.avi", "*.mov"]:
                examples.extend([str(p) for p in example_dir.glob(ext)])

        if examples:
            gr.Examples(
                examples=[[ex, 0.85, "all"] for ex in examples[:3]],
                inputs=[video_input, fall_threshold, viz_keypoints],
                outputs=[video_output, prob_graph, result_text],
                fn=process_video,
                cache_examples=False,
            )

        # 이벀트 μ—°κ²°
        submit_btn.click(
            fn=process_video,
            inputs=[video_input, fall_threshold, viz_keypoints],
            outputs=[video_output, prob_graph, result_text],
        )

        # ν‘Έν„°
        gr.Markdown(
            """
            ---
            **References:**
            - [YOLOv11](https://github.com/ultralytics/ultralytics) - Pose Estimation
            - [ST-GCN](https://arxiv.org/abs/1801.07455) - Spatial Temporal Graph Convolutional Networks
            - AI Hub Fall Detection Dataset
            """
        )

    return demo


# -----------------------------------------------------------------------------
# 메인 μ‹€ν–‰
# -----------------------------------------------------------------------------
if __name__ == "__main__":
    demo = create_demo()
    demo.queue(max_size=10).launch(
        server_name="0.0.0.0",
        server_port=7860,
        share=True,
        show_error=True,
    )