twitch-highlight-detector/scene_detector.py

#!/opt/vlm_env/bin/python3
"""
SCENE DETECTION + CLASSIFICATION (GPU Accelerated)
Detecta cambios de escena con FFmpeg (rápido) y clasifica cada una
Compatible con RX 6800 XT (16GB VRAM)
"""

import subprocess
import json
import re
from pathlib import Path

print("=" * 70)
print("🎬 SCENE DETECTOR + CLASSIFIER")
print("=" * 70)
print("Paso 1: Detectar cambios de escena con FFmpeg (rápido)")
print("Paso 2: Clasificar cada escena (gameplay vs hablando)")
print()

video_path = (
    "/home/ren/proyectos/editor/twitch-highlight-detector/nuevo_stream_360p.mp4"
)

# PASO 1: Detectar cambios de escena (threshold 0.3 = cambios significativos)
print("🔍 Detectando cambios de escena...")
result = subprocess.run(
    [
        "ffmpeg",
        "-i",
        video_path,
        "-vf",
        "select=gt(scene\,0.3),showinfo",
        "-f",
        "null",
        "-",
    ],
    capture_output=True,
    text=True,
)

# Extraer timestamps de cambios de escena
scene_changes = []
for line in result.stderr.split("\n"):
    if "pts_time:" in line:
        match = re.search(r"pts_time:(\d+\.\d+)", line)
        if match:
            ts = float(match.group(1))
            if ts > 455:  # Saltar intro
                scene_changes.append(ts)

print(f"✅ {len(scene_changes)} cambios de escena detectados")

# PASO 2: Analizar transcripción en cada escena
print("\n📊 Analizando contenido de cada escena...")

with open(
    "/home/ren/proyectos/editor/twitch-highlight-detector/transcripcion_rage.json", "r"
) as f:
    trans = json.load(f)

# Crear segmentos entre cambios de escena
segments = []
prev_ts = 455

for ts in sorted(scene_changes):
    if ts - prev_ts > 30:  # Mínimo 30 segundos
        segments.append({"start": prev_ts, "end": ts, "duration": ts - prev_ts})
    prev_ts = ts

# Agregar último segmento
result = subprocess.run(
    [
        "ffprobe",
        "-v",
        "error",
        "-show_entries",
        "format=duration",
        "-of",
        "default=noprint_wrappers=1:nokey=1",
        video_path,
    ],
    capture_output=True,
    text=True,
)
duration = float(result.stdout.strip())

if duration - prev_ts > 30:
    segments.append({"start": prev_ts, "end": duration, "duration": duration - prev_ts})

print(f"✅ {len(segments)} segmentos para analizar")

# PASO 3: Clasificar cada segmento usando transcripción
print("\n🎯 Clasificando segmentos (gameplay vs hablando)...")

for seg in segments:
    # Buscar transcripción en este rango
    seg_text = []
    rage_score = 0

    for t in trans["segments"]:
        if seg["start"] <= t["start"] <= seg["end"]:
            seg_text.append(t["text"].lower())

            # Calcular score de rage
            if any(word in t["text"].lower() for word in ["puta", "mierda", "joder"]):
                rage_score += 10
            elif any(
                word in t["text"].lower() for word in ["me mataron", "kill", "muere"]
            ):
                rage_score += 8
            elif any(word in t["text"].lower() for word in ["ulti", "flash", "gank"]):
                rage_score += 5

    full_text = " ".join(seg_text)

    # Clasificar
    if any(
        word in full_text for word in ["seleccion", "champions", "ban", "pick", "elij"]
    ):
        seg["type"] = "SELECCION"
        seg["keep"] = False
    elif any(
        word in full_text for word in ["cuento", "historia", "ayer", "comida", "vida"]
    ):
        seg["type"] = "HABLANDO"
        seg["keep"] = False
    elif rage_score >= 5 or any(
        word in full_text for word in ["kill", "matan", "pelea", "fight"]
    ):
        seg["type"] = "GAMEPLAY"
        seg["keep"] = True
        seg["rage_score"] = rage_score
    else:
        seg["type"] = "GAMEPLAY_NEUTRO"
        seg["keep"] = True
        seg["rage_score"] = rage_score

# Mostrar resultados
print("\n" + "=" * 70)
print("SEGMENTOS CLASIFICADOS")
print("=" * 70)

gameplay_segments = [s for s in segments if s["keep"]]

for i, seg in enumerate(segments, 1):
    mins_s, secs_s = divmod(int(seg["start"]), 60)
    mins_e, secs_e = divmod(int(seg["end"]), 60)
    icon = "✅" if seg["keep"] else "❌"
    print(
        f"{icon} {i}. {mins_s:02d}:{secs_s:02d} - {mins_e:02d}:{secs_e:02d} "
        f"({seg['duration'] // 60:.0f}m) [{seg['type']}]"
    )
    if seg.get("rage_score"):
        print(f"     Rage score: {seg['rage_score']}")

print(f"\n{'=' * 70}")
print(f"RESUMEN")
print(f"{'=' * 70}")
print(f"Total segmentos: {len(segments)}")
print(f"Gameplay útil: {len(gameplay_segments)}")
total_gameplay = sum(s["duration"] for s in gameplay_segments)
print(f"Tiempo gameplay: {total_gameplay // 60:.0f}m {total_gameplay % 60:.0f}s")

# Guardar gameplay útil
with open(
    "/home/ren/proyectos/editor/twitch-highlight-detector/gameplay_scenes.json", "w"
) as f:
    json.dump(gameplay_segments, f, indent=2)

print(f"\n💾 Guardado: gameplay_scenes.json")
print("\nAhora extrae highlights SOLO de estos rangos confirmados.")