import cv2 import numpy as np import mediapipe as mp from advanced_fitter import fitter class PerfectFitModel: def __init__(self): self.mp_pose = mp.solutions.pose self.pose = self.mp_pose.Pose( static_image_mode=True, model_complexity=2, min_detection_confidence=0.7 ) def predict(self, person_image, clothing_path, clothing_type, size): """HF Spaces prediction method""" rgb_image = cv2.cvtColor(person_image, cv2.COLOR_BGR2RGB) results = self.pose.process(rgb_image) if not results.pose_landmarks: return person_image h, w = person_image.shape[:2] keypoints = [(lm.x * w, lm.y * h) for lm in results.pose_landmarks.landmark] clothing_img = cv2.imread(clothing_path, cv2.IMREAD_UNCHANGED) measurements = fitter.extract_body_measurements(keypoints) warped_clothing, _ = fitter.warp_clothing_to_perfect_fit( clothing_img, measurements, clothing_type ) # Apply perfect overlay (simplified for HF) result = self._smart_overlay(person_image, warped_clothing, keypoints, clothing_type) return result def _smart_overlay(self, person_img, clothing, keypoints, clothing_type): # Simplified overlay logic for speed LEFT_SHOULDER, RIGHT_SHOULDER, LEFT_HIP = 11, 12, 23 x1 = int(min(kp[0] for kp in keypoints[:33]) - 30) y1 = int(min(kp[1] for kp in keypoints[:33]) - 50) x2 = int(max(kp[0] for kp in keypoints[:33]) + 30) y2 = int(keypoints[LEFT_HIP][1] + 40) x1, y1 = max(0, x1), max(0, y1) x2, y2 = min(person_img.shape[1], x2), min(person_img.shape[0], y2) region = person_img[y1:y2, x1:x2] clothing_resized = cv2.resize(clothing, (region.shape[1], region.shape[0])) if len(clothing_resized.shape) == 3 and clothing_resized.shape[2] == 4: alpha = clothing_resized[:, :, 3] / 255.0 for c in range(3): region[:, :, c] = region[:, :, c] * (1 - alpha) + clothing_resized[:, :, c] * alpha person_img[y1:y2, x1:x2] = region return person_img