Create processing/video/video_processor.py
Browse files
processing/video/video_processor.py
ADDED
|
@@ -0,0 +1,121 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/usr/bin/env python3
|
| 2 |
+
"""
|
| 3 |
+
Compatibility shim: CoreVideoProcessor
|
| 4 |
+
|
| 5 |
+
Bridges the legacy import `from processing.video.video_processor import CoreVideoProcessor`
|
| 6 |
+
to the modern pipeline functions living in `utils.cv_processing` and models in `core.models`.
|
| 7 |
+
"""
|
| 8 |
+
|
| 9 |
+
from __future__ import annotations
|
| 10 |
+
|
| 11 |
+
from dataclasses import dataclass
|
| 12 |
+
from typing import Optional, Dict, Any, Tuple
|
| 13 |
+
|
| 14 |
+
import cv2
|
| 15 |
+
import numpy as np
|
| 16 |
+
|
| 17 |
+
from utils.logger import get_logger
|
| 18 |
+
from core.models import ModelManager
|
| 19 |
+
# ← these funcs are the ones you showed (in utils/cv_processing.py)
|
| 20 |
+
from utils.cv_processing import (
|
| 21 |
+
segment_person_hq,
|
| 22 |
+
refine_mask_hq,
|
| 23 |
+
replace_background_hq,
|
| 24 |
+
create_professional_background,
|
| 25 |
+
validate_video_file,
|
| 26 |
+
)
|
| 27 |
+
|
| 28 |
+
@dataclass
|
| 29 |
+
class ProcessorConfig:
|
| 30 |
+
background_preset: str = "minimalist" # key in PROFESSIONAL_BACKGROUNDS
|
| 31 |
+
write_fps: Optional[float] = None # None -> keep source fps
|
| 32 |
+
|
| 33 |
+
|
| 34 |
+
class CoreVideoProcessor:
|
| 35 |
+
"""
|
| 36 |
+
Minimal, safe implementation used by core/app.py.
|
| 37 |
+
It relies on ModelManager (SAM2 + MatAnyone) and your cv_processing helpers.
|
| 38 |
+
"""
|
| 39 |
+
|
| 40 |
+
def __init__(self, config: Optional[ProcessorConfig] = None, models: Optional[ModelManager] = None):
|
| 41 |
+
self.log = get_logger(f"{__name__}.CoreVideoProcessor")
|
| 42 |
+
self.config = config or ProcessorConfig()
|
| 43 |
+
self.models = models or ModelManager()
|
| 44 |
+
try:
|
| 45 |
+
self.models.load_all()
|
| 46 |
+
except Exception as e:
|
| 47 |
+
self.log.warning(f"Model load issue (will use fallbacks if needed): {e}")
|
| 48 |
+
|
| 49 |
+
# --- single-frame API (useful for images or per-frame video loop) ---
|
| 50 |
+
def process_frame(self, frame: np.ndarray, background: np.ndarray) -> Dict[str, Any]:
|
| 51 |
+
"""Return dict with composited frame + mask; always succeeds with fallbacks."""
|
| 52 |
+
predictor = None
|
| 53 |
+
try:
|
| 54 |
+
predictor = self.models.get_sam2().predictor
|
| 55 |
+
except Exception as e:
|
| 56 |
+
self.log.warning(f"SAM2 predictor unavailable, using fallback: {e}")
|
| 57 |
+
|
| 58 |
+
# 1) segment
|
| 59 |
+
mask = segment_person_hq(frame, predictor, fallback_enabled=True)
|
| 60 |
+
|
| 61 |
+
# 2) refine
|
| 62 |
+
matanyone = None
|
| 63 |
+
try:
|
| 64 |
+
matanyone = self.models.get_matanyone()
|
| 65 |
+
except Exception as e:
|
| 66 |
+
self.log.warning(f"MatAnyone unavailable, using OpenCV refinement: {e}")
|
| 67 |
+
|
| 68 |
+
mask_refined = refine_mask_hq(frame, mask, matanyone, fallback_enabled=True)
|
| 69 |
+
|
| 70 |
+
# 3) composite
|
| 71 |
+
out = replace_background_hq(frame, mask_refined, background, fallback_enabled=True)
|
| 72 |
+
|
| 73 |
+
return {"frame": out, "mask": mask_refined}
|
| 74 |
+
|
| 75 |
+
# --- simple video API (covers typical usage in older core/app.py code) ---
|
| 76 |
+
def process_video(self, input_path: str, output_path: str, bg_config: Optional[Dict[str, Any]] = None) -> Dict[str, Any]:
|
| 77 |
+
"""Process a full video; returns basic stats."""
|
| 78 |
+
ok, msg = validate_video_file(input_path)
|
| 79 |
+
if not ok:
|
| 80 |
+
raise ValueError(f"Invalid video: {msg}")
|
| 81 |
+
self.log.info(f"Video validation: {msg}")
|
| 82 |
+
|
| 83 |
+
cap = cv2.VideoCapture(input_path)
|
| 84 |
+
if not cap.isOpened():
|
| 85 |
+
raise RuntimeError(f"Could not open video: {input_path}")
|
| 86 |
+
|
| 87 |
+
width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
|
| 88 |
+
height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
|
| 89 |
+
fps = cap.get(cv2.CAP_PROP_FPS)
|
| 90 |
+
fps_out = self.config.write_fps or (fps if fps and fps > 0 else 25.0)
|
| 91 |
+
|
| 92 |
+
fourcc = cv2.VideoWriter_fourcc(*"mp4v")
|
| 93 |
+
writer = cv2.VideoWriter(output_path, fourcc, fps_out, (width, height))
|
| 94 |
+
|
| 95 |
+
# Build background (once)
|
| 96 |
+
from utils.cv_processing import PROFESSIONAL_BACKGROUNDS # local import to avoid circulars
|
| 97 |
+
preset = self.config.background_preset
|
| 98 |
+
cfg = bg_config or PROFESSIONAL_BACKGROUNDS.get(preset, PROFESSIONAL_BACKGROUNDS["minimalist"])
|
| 99 |
+
background = create_professional_background(cfg, width, height)
|
| 100 |
+
|
| 101 |
+
frame_count = 0
|
| 102 |
+
try:
|
| 103 |
+
while True:
|
| 104 |
+
ret, frame = cap.read()
|
| 105 |
+
if not ret:
|
| 106 |
+
break
|
| 107 |
+
res = self.process_frame(frame, background)
|
| 108 |
+
writer.write(res["frame"])
|
| 109 |
+
frame_count += 1
|
| 110 |
+
finally:
|
| 111 |
+
cap.release()
|
| 112 |
+
writer.release()
|
| 113 |
+
|
| 114 |
+
self.log.info(f"Processed {frame_count} frames → {output_path}")
|
| 115 |
+
return {"frames": frame_count, "width": width, "height": height, "fps_out": fps_out}
|
| 116 |
+
|
| 117 |
+
# Backward-compat export name (if someone expects `VideoProcessor`)
|
| 118 |
+
try:
|
| 119 |
+
VideoProcessor
|
| 120 |
+
except NameError:
|
| 121 |
+
VideoProcessor = CoreVideoProcessor
|