更新

models/LatentSync/scripts/server.py

@@ -65,14 +65,15 @@ async def lifespan(app: FastAPI):
     # --- 模型加载逻辑 (参考 inference.py) ---
     print("⏳ 正在加载 LatentSync 模型...")
     
-    # 默认配置路径 (相对于根目录)
-    unet_config_path = "configs/unet/stage2_512.yaml"
-    ckpt_path = "checkpoints/latentsync_unet.pt"
+    # 使用绝对路径，确保可以从任意目录启动
+    latentsync_root = Path(__file__).resolve().parent.parent  # scripts -> LatentSync 根目录
+    unet_config_path = latentsync_root / "configs" / "unet" / "stage2_512.yaml"
+    ckpt_path = latentsync_root / "checkpoints" / "latentsync_unet.pt"
     
-    if not os.path.exists(unet_config_path):
-        print(f"⚠️ 找不到配置文件: {unet_config_path}，请确保在 models/LatentSync 根目录运行")
+    if not unet_config_path.exists():
+        print(f"⚠️ 找不到配置文件: {unet_config_path}")
     
-    config = OmegaConf.load(unet_config_path)
+    config = OmegaConf.load(str(unet_config_path))
     
     # Check GPU
     is_fp16_supported = torch.cuda.is_available() and torch.cuda.get_device_capability()[0] > 7
@@ -85,13 +86,13 @@ async def lifespan(app: FastAPI):
     else:
         print("⚠️  警告: 未检测到 GPU，将使用 CPU 进行推理 (速度极慢)")
     
-    scheduler = DDIMScheduler.from_pretrained("configs")
+    scheduler = DDIMScheduler.from_pretrained(str(latentsync_root / "configs"))
     
     # Whisper Model
     if config.model.cross_attention_dim == 768:
-        whisper_path = "checkpoints/whisper/small.pt"
+        whisper_path = str(latentsync_root / "checkpoints" / "whisper" / "small.pt")
     else:
-        whisper_path = "checkpoints/whisper/tiny.pt"
+        whisper_path = str(latentsync_root / "checkpoints" / "whisper" / "tiny.pt")
         
     audio_encoder = Audio2Feature(
         model_path=whisper_path,
@@ -108,7 +109,7 @@ async def lifespan(app: FastAPI):
     # UNet
     unet, _ = UNet3DConditionModel.from_pretrained(
         OmegaConf.to_container(config.model),
-        ckpt_path,
+        str(ckpt_path),
         device="cpu", # Load to CPU first to save memory during init
     )
     unet = unet.to(dtype=dtype)
@@ -129,6 +130,7 @@ async def lifespan(app: FastAPI):
     models["pipeline"] = pipeline
     models["config"] = config
     models["dtype"] = dtype
+    models["latentsync_root"] = latentsync_root
     
     print("✅ LatentSync 模型加载完成，服务就绪！")
     yield
@@ -167,6 +169,7 @@ async def generate_lipsync(req: LipSyncRequest):
         pipeline = models["pipeline"]
         config = models["config"]
         dtype = models["dtype"]
+        latentsync_root = models["latentsync_root"]
         
         # Set seed
         if req.seed != -1:
@@ -185,7 +188,7 @@ async def generate_lipsync(req: LipSyncRequest):
             weight_dtype=dtype,
             width=config.data.resolution,
             height=config.data.resolution,
-            mask_image_path=config.data.mask_image_path,
+            mask_image_path=str(latentsync_root / config.data.mask_image_path),
             temp_dir=req.temp_dir,
         )