MogensR's picture
Update utils/refinement/mask_refiner.py
62df638
"""
Complete utils/__init__.py with all required functions
Provides direct implementations to avoid import recursion
"""
import cv2
import numpy as np
from PIL import Image
import torch
import logging
from typing import Optional, Tuple, Dict, Any, List
import tempfile
import os
logger = logging.getLogger(__name__)
# Professional backgrounds configuration
PROFESSIONAL_BACKGROUNDS = {
"office": {"color": (240, 248, 255), "gradient": True},
"studio": {"color": (32, 32, 32), "gradient": False},
"nature": {"color": (34, 139, 34), "gradient": True},
"abstract": {"color": (75, 0, 130), "gradient": True},
"white": {"color": (255, 255, 255), "gradient": False},
"black": {"color": (0, 0, 0), "gradient": False}
}
def validate_video_file(video_path: str) -> bool:
"""Validate if video file is readable"""
try:
if not os.path.exists(video_path):
return False
cap = cv2.VideoCapture(video_path)
if not cap.isOpened():
return False
ret, frame = cap.read()
cap.release()
return ret and frame is not None
except Exception as e:
logger.error(f"Video validation failed: {e}")
return False
def segment_person_hq(frame: np.ndarray, use_sam2: bool = True) -> Optional[np.ndarray]:
"""High-quality person segmentation using SAM2 or fallback methods"""
try:
if use_sam2:
# Try SAM2 segmentation
try:
from sam2.sam2_image_predictor import SAM2ImagePredictor
from sam2.build_sam import build_sam2
from huggingface_hub import hf_hub_download
# Load SAM2 model
sam_checkpoint = hf_hub_download("facebook/sam2-hiera-base-plus", "sam2_hiera_b+.pt")
sam_model = build_sam2(model_name='sam2_hiera_base_plus_t', ckpt_path=sam_checkpoint)
predictor = SAM2ImagePredictor(sam_model)
# Set image and predict
predictor.set_image(frame)
# Use center point as prompt (assuming person is in center)
h, w = frame.shape[:2]
center_point = np.array([[w//2, h//2]])
center_label = np.array([1])
masks, scores, _ = predictor.predict(
point_coords=center_point,
point_labels=center_label,
multimask_output=False
)
return masks[0] if len(masks) > 0 else None
except Exception as e:
logger.warning(f"SAM2 segmentation failed: {e}, falling back to simple method")
# Fallback: Simple person detection using background subtraction
return _simple_person_segmentation(frame)
except Exception as e:
logger.error(f"Person segmentation failed: {e}")
return None
def _simple_person_segmentation(frame: np.ndarray) -> np.ndarray:
"""Simple person segmentation using color-based methods"""
# Convert to HSV for better color detection
hsv = cv2.cvtColor(frame, cv2.COLOR_RGB2HSV)
# Create mask for common background colors (green screen, white, etc.)
# Green screen detection
lower_green = np.array([40, 40, 40])
upper_green = np.array([80, 255, 255])
green_mask = cv2.inRange(hsv, lower_green, upper_green)
# White background detection
lower_white = np.array([0, 0, 200])
upper_white = np.array([180, 30, 255])
white_mask = cv2.inRange(hsv, lower_white, upper_white)
# Combine masks
bg_mask = cv2.bitwise_or(green_mask, white_mask)
# Invert to get person mask
person_mask = cv2.bitwise_not(bg_mask)
# Clean up mask with morphological operations
kernel = np.ones((5, 5), np.uint8)
person_mask = cv2.morphologyEx(person_mask, cv2.MORPH_CLOSE, kernel)
person_mask = cv2.morphologyEx(person_mask, cv2.MORPH_OPEN, kernel)
# Convert to float and normalize
return person_mask.astype(np.float32) / 255.0
def refine_mask_hq(mask: np.ndarray, frame: np.ndarray, use_matanyone: bool = True) -> np.ndarray:
"""High-quality mask refinement using MatAnyone or fallback methods"""
try:
if use_matanyone:
try:
from matanyone import InferenceCore
# Initialize MatAnyone
device = "cuda" if torch.cuda.is_available() else "cpu"
processor = InferenceCore(model_name="PeiqingYang/MatAnyone-v1.0", device=device)
# Convert inputs to PIL Images
frame_pil = Image.fromarray(frame.astype(np.uint8))
mask_pil = Image.fromarray((mask * 255).astype(np.uint8))
# Refine mask
refined_mask = processor.infer(frame_pil, mask_pil)
# Convert back to numpy
return np.array(refined_mask).astype(np.float32) / 255.0
except Exception as e:
logger.warning(f"MatAnyone refinement failed: {e}, using simple refinement")
# Fallback: Simple mask refinement
return _simple_mask_refinement(mask, frame)
except Exception as e:
logger.error(f"Mask refinement failed: {e}")
return mask
def _simple_mask_refinement(mask: np.ndarray, frame: np.ndarray) -> np.ndarray:
"""Simple mask refinement using OpenCV operations"""
# Convert mask to uint8
mask_uint8 = (mask * 255).astype(np.uint8)
# Apply Gaussian blur for smoother edges
mask_blurred = cv2.GaussianBlur(mask_uint8, (5, 5), 0)
# Apply bilateral filter to preserve edges while smoothing
mask_refined = cv2.bilateralFilter(mask_blurred, 9, 75, 75)
# Convert back to float
return mask_refined.astype(np.float32) / 255.0
def replace_background_hq(frame: np.ndarray, mask: np.ndarray, background: np.ndarray) -> np.ndarray:
"""High-quality background replacement with proper compositing"""
try:
# Ensure all inputs are the same size
h, w = frame.shape[:2]
background_resized = cv2.resize(background, (w, h))
# Ensure mask has 3 channels
if len(mask.shape) == 2:
mask_3d = np.stack([mask] * 3, axis=-1)
else:
mask_3d = mask
# Apply feathering to mask edges for smoother blending
mask_feathered = _apply_feathering(mask_3d)
# Composite the image
result = frame * mask_feathered + background_resized * (1 - mask_feathered)
return result.astype(np.uint8)
except Exception as e:
logger.error(f"Background replacement failed: {e}")
return frame
def _apply_feathering(mask: np.ndarray, feather_amount: int = 3) -> np.ndarray:
"""Apply feathering to mask edges for smoother blending"""
if len(mask.shape) == 3:
# Work with single channel
mask_single = mask[:, :, 0]
else:
mask_single = mask
# Apply Gaussian blur for feathering
mask_feathered = cv2.GaussianBlur(mask_single, (feather_amount*2+1, feather_amount*2+1), 0)
# Restore 3 channels if needed
if len(mask.shape) == 3:
mask_feathered = np.stack([mask_feathered] * 3, axis=-1)
return mask_feathered
def create_professional_background(bg_type: str, width: int, height: int) -> np.ndarray:
"""Create professional background of specified type and size"""
try:
if bg_type not in PROFESSIONAL_BACKGROUNDS:
bg_type = "office" # Default fallback
config = PROFESSIONAL_BACKGROUNDS[bg_type]
color = config["color"]
use_gradient = config["gradient"]
if use_gradient:
# Create gradient background
background = _create_gradient_background(color, width, height)
else:
# Create solid color background
background = np.full((height, width, 3), color, dtype=np.uint8)
return background
except Exception as e:
logger.error(f"Background creation failed: {e}")
# Return white background as fallback
return np.full((height, width, 3), (255, 255, 255), dtype=np.uint8)
def _create_gradient_background(base_color: Tuple[int, int, int], width: int, height: int) -> np.ndarray:
"""Create a gradient background from base color"""
# Create gradient from darker to lighter version of base color
r, g, b = base_color
# Create darker version (multiply by 0.7)
dark_color = (int(r * 0.7), int(g * 0.7), int(b * 0.7))
# Create gradient
background = np.zeros((height, width, 3), dtype=np.uint8)
for y in range(height):
# Calculate blend factor (0 to 1)
blend = y / height
# Interpolate between dark and light color
current_r = int(dark_color[0] * (1 - blend) + r * blend)
current_g = int(dark_color[1] * (1 - blend) + g * blend)
current_b = int(dark_color[2] * (1 - blend) + b * blend)
background[y, :] = [current_r, current_g, current_b]
return background
# Export all functions
__all__ = [
"segment_person_hq",
"refine_mask_hq",
"replace_background_hq",
"create_professional_background",
"PROFESSIONAL_BACKGROUNDS",
"validate_video_file"
]