feat/igpu-ov: OpenVINO iGPU detector (--gpu igpu) on Intel Level Zero

- ov_detector.OVFaceDetector: OV inference + ultralytics letterbox(1280 sq)
  + greedy NMS; model.predict()-compatible shim, process_video() unchanged
- anime_facesaver.py: --gpu igpu branch; CUDA path untouched (moved to else)
- CUDA path untouched; main() unchanged
- validate_ov.py: cross-check vs torch CPU baseline
  measured 2026-10-20 i3-N305: 59/59 matched, mean IoU 0.987,
  only_base=0 only_ov=0; 5.0s/img GPU vs 14.5s/img CPU OpenVINO-CPU vs
  torch-CPU baseline (25 real stills, yolov8x6_animeface)

anime_facesaver.py

@@ -161,6 +161,10 @@ def resolve_gpu(gpu_arg):
     if gpu_arg.lower() == "cpu":
         os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
         return "cpu"
+    if gpu_arg.lower() == "igpu":
+        # Intel iGPU via OpenVINO (Level Zero) — no CUDA, no torch pinning.
+        # OVFaceDetector takes over model loading in main().
+        return "igpu"
     if gpu_arg.lower() == "auto":
         return None  # let ultralytics/torch decide
     os.environ["CUDA_VISIBLE_DEVICES"] = gpu_arg
@@ -353,16 +357,32 @@ def main():
     from ultralytics import YOLO
     import cv2
 
-    if device == "cuda:0" and not torch.cuda.is_available():
-        print("CUDA requested but not visible after pinning "
-              f"(--gpu {args.gpu}). Falling back to CPU.")
-        device = "cpu"
-    print(f"Device: {device}"
-          + (f" (physical GPU {args.gpu})" if device == "cuda" else ""))
+    if device == "igpu":
+        # Intel iGPU path: OpenVINO replaces ultralytics' torch predict loop.
+        # Ultralytics import stays (shared pre/post code paths), but weights
+        # load as ONNX through OVFaceDetector instead of torch.
+        from ov_detector import OVFaceDetector
+        onnx_path = os.path.splitext(args.model)[0] + ".onnx"
+        if not os.path.exists(onnx_path):
+            print(f"Error: --gpu igpu needs an exported ONNX at {onnx_path}\n"
+                  "  (one-time: YOLO('<pt>').export(format='onnx', "
+                  "dynamic=True, simplify=True, opset=17))")
+            sys.exit(1)
+        model = OVFaceDetector(onnx_path, device="GPU", imgsz=args.imgsz)
+        names = model.names
+        print("Device: igpu (OpenVINO, Intel iGPU via Level Zero)")
+        print(f"Model loaded (OpenVINO GPU). Classes: {names}")
+    else:
+        if device == "cuda:0" and not torch.cuda.is_available():
+            print("CUDA requested but not visible after pinning "
+                  f"(--gpu {args.gpu}). Falling back to CPU.")
+            device = "cpu"
+        print(f"Device: {device}"
+              + (f" (physical GPU {args.gpu})" if device == "cuda" else ""))
 
-    model = YOLO(args.model)
-    names = model.names
-    print(f"Model loaded. Classes: {names}")
+        model = YOLO(args.model)
+        names = model.names
+        print(f"Model loaded. Classes: {names}")
 
     def draw(img, boxes):
         for (a, b, c, d) in boxes.int().tolist():