初始化：换发型/换发色/训练发型服务

包含：
- hair_service_sd: 主服务（换发型/换发色/生发，端口8801）
- photo_service: LoRA调度+训练（端口32678）
- hair_grow_service: 调试测试页（端口8888，含4个测试页）
- 批量训练脚本（batch_train_hairstyles.py）
- 发际线mask自动识别（hairline_mask.py，4种方案）
- 手绘mask换发型（hair_swap_manual.py）
- 文档：README.md + LARGE_FILES.md + docs/

大文件（模型权重200G、训练数据123G）已排除，见 LARGE_FILES.md
OSS/COS密钥已脱敏为环境变量，原文件备份在本地

project/hair_service_sd/models/Generator_BaldSeg.py

@@ -0,0 +1,69 @@
+import os
+import time
+import torch
+import torch.nn.parallel
+
+import numpy as np
+from utils import landmark_processor
+import cv2
+
+# modelRoot = "/home/yangchaojie/Desktop/hairstyle/hairstyle_infer/weights"
+modelRoot = "./weights"
+
+label_map = [
+    [0, 0, 0],
+    [255, 0, 0],
+    [0, 0, 255],
+    [0, 255, 0],
+    [0, 255, 255],
+    # [0, 255, 0]
+]
+class Generator_BaldSeg_5c(object):
+    def __init__(self, gpu_flag, gpu_id):
+
+        if not gpu_flag:
+            self.device = torch.device("cpu")
+        else:
+            self.device = torch.device('cuda:{0}'.format(gpu_id))
+
+        # load seg model
+        self.model_dir = modelRoot
+        self.pre_trained_model = os.path.join(self.model_dir, "ori_hair_checkpoint_7660_0611.pt")
+        self.net = torch.jit.load(self.pre_trained_model, map_location=self.device).to(self.device)
+        self.net.eval()
+
+        self.output_img_size = 512
+        self.img_ratio = 0.4
+
+    def label_to_mask(self, label_np):
+        label_np = label_np.astype(np.int32)[:, :, np.newaxis]
+        mask = np.zeros((label_np.shape[0], label_np.shape[1], 3), dtype=np.uint8)
+        for id, color in enumerate(label_map):
+            index = (label_np == id).all(axis=2)
+            mask[index] = color
+        return mask
+    def forward(self, image, alpha, landmarks1k):
+        image_to_face_mat = landmark_processor.get_transform_mat_full_face_ratio_deeplab(landmarks1k, self.output_img_size, self.img_ratio)
+        img_ = (image * (1 - alpha[:, :, np.newaxis].astype(np.float32) / 255)).astype(np.uint8)
+        img_cuted = cv2.warpAffine(img_, image_to_face_mat, (self.output_img_size, self.output_img_size), flags=cv2.INTER_LANCZOS4, borderMode=cv2.BORDER_CONSTANT, borderValue=[255, 255, 255])
+        inv_image_to_face_mat = cv2.invertAffineTransform(image_to_face_mat)
+
+        img = (img_cuted.astype(np.float32) / 255).transpose((2, 0, 1))
+        img = np.expand_dims(img, axis=0)
+        img = torch.from_numpy(img).to(self.device)  #cuda(self.gpu_id)
+
+        with torch.no_grad():
+            output = self.net(img)
+
+        pred = output.detach().cpu().numpy().squeeze().astype(np.float32) #torch.max(output[:1], 1)[1].detach().cpu().numpy().squeeze().astype(np.float32)
+        mask = self.label_to_mask(pred)
+
+        # cv2.imshow("img_cuted: ", img_cuted)
+        # cv2.imshow("mask: ", mask)
+        # cv2.waitKey()
+
+        black_img = np.zeros(image.shape).astype(np.uint8)
+        cv2.warpAffine(mask, inv_image_to_face_mat, (image.shape[1], image.shape[0]),
+                       dst=black_img, flags=cv2.INTER_NEAREST, borderMode=cv2.BORDER_TRANSPARENT)
+
+        return black_img