Merge pull request #146 from SharpAI/develop

Develop

Author: solderzzc

skills/detection/yolo-detection-2026/deploy.bat

@@ -120,9 +120,13 @@ if exist "%SKILL_DIR%\scripts\env_config.py" (
 if defined ENV_CONFIG_DIR (
     echo %LOG_PREFIX% Detecting hardware via env_config.py...>&2
 
-    REM Run env_config detection via Python
-    for /f "tokens=*" %%B in ('"%VPYTHON%" -c "import sys; sys.path.insert(0, r'!ENV_CONFIG_DIR!'); from env_config import HardwareEnv; env = HardwareEnv.detect(); print(env.backend)" 2^>nul') do (
-        set "DETECTED_BACKEND=%%B"
+    REM Run env_config detection via Python (temp file avoids for /f quoting issues)
+    set "ENV_CONFIG_DIR=!ENV_CONFIG_DIR!"
+    set "_TMPBACK=%TEMP%\_aegis_detect_backend.txt"
+    "%VPYTHON%" -c "import sys,os; sys.path.insert(0, os.environ['ENV_CONFIG_DIR']); from env_config import HardwareEnv; env = HardwareEnv.detect(); print(env.backend)" > "!_TMPBACK!" 2>nul
+    if exist "!_TMPBACK!" (
+        set /p DETECTED_BACKEND=<"!_TMPBACK!"
+        del "!_TMPBACK!" 2>nul
     )
 
     REM Validate backend value (Windows: only cuda, intel, cpu are realistic)
@@ -187,7 +191,7 @@ if "!BACKEND!" neq "cpu" (
     echo %LOG_PREFIX% Pre-converting model to optimized format for !BACKEND!...>&2
     echo {"event": "progress", "stage": "optimize", "message": "Converting model for !BACKEND! (~30-120s)..."}
 
-    "%VPYTHON%" -c "import sys; sys.path.insert(0, r'!ENV_CONFIG_DIR!'); from env_config import HardwareEnv; env = HardwareEnv.detect(); from ultralytics import YOLO; model = YOLO('yolo26n.pt'); result = env.export_model(model, 'yolo26n'); print(f'Optimized model exported: {result}' if result else 'Export skipped or failed')" 2>&1
+    "%VPYTHON%" -c "import sys,os; sys.path.insert(0, os.environ['ENV_CONFIG_DIR']); from env_config import HardwareEnv; env = HardwareEnv.detect(); from ultralytics import YOLO; model = YOLO('yolo26n.pt'); result = env.export_model(model, 'yolo26n'); print(f'Optimized model exported: {result}' if result else 'Export skipped or failed')" 2>&1
 
     if !errorlevel! equ 0 (
         echo {"event": "progress", "stage": "optimize", "message": "Model optimization complete"}
@@ -199,7 +203,7 @@ if "!BACKEND!" neq "cpu" (
     echo %LOG_PREFIX% Pre-converting model to ONNX for CPU...>&2
     echo {"event": "progress", "stage": "optimize", "message": "Converting model for cpu (~30-120s)..."}
 
-    "%VPYTHON%" -c "import sys; sys.path.insert(0, r'!ENV_CONFIG_DIR!'); from env_config import HardwareEnv; env = HardwareEnv.detect(); from ultralytics import YOLO; model = YOLO('yolo26n.pt'); result = env.export_model(model, 'yolo26n'); print(f'Optimized model exported: {result}' if result else 'Export skipped or failed')" 2>&1
+    "%VPYTHON%" -c "import sys,os; sys.path.insert(0, os.environ['ENV_CONFIG_DIR']); from env_config import HardwareEnv; env = HardwareEnv.detect(); from ultralytics import YOLO; model = YOLO('yolo26n.pt'); result = env.export_model(model, 'yolo26n'); print(f'Optimized model exported: {result}' if result else 'Export skipped or failed')" 2>&1
 
     if !errorlevel! equ 0 (
         echo {"event": "progress", "stage": "optimize", "message": "Model optimization complete"}
@@ -212,7 +216,7 @@ if "!BACKEND!" neq "cpu" (
 REM ─── Step 6: Verify installation ───────────────────────────────────────────
 
 echo %LOG_PREFIX% Verifying installation...>&2
-"%VPYTHON%" -c "import sys, json; sys.path.insert(0, r'!ENV_CONFIG_DIR!'); from env_config import HardwareEnv; env = HardwareEnv.detect(); print(json.dumps(env.to_dict(), indent=2))" 2>&1
+"%VPYTHON%" -c "import sys,os,json; sys.path.insert(0, os.environ['ENV_CONFIG_DIR']); from env_config import HardwareEnv; env = HardwareEnv.detect(); print(json.dumps(env.to_dict(), indent=2))" 2>&1
 
 echo {"event": "complete", "backend": "!BACKEND!", "message": "YOLO 2026 skill installed (!BACKEND! backend)"}
 echo %LOG_PREFIX% Done! Backend: !BACKEND!>&2

skills/detection/yolo-detection-2026/scripts/detect.py

@@ -240,14 +240,22 @@ def main():
         target_classes = [c.strip() for c in target_classes.split(",")]
 
     # ── Hardware detection & optimized model loading ──
+    emit({"event": "progress", "stage": "init", "message": "Detecting compute hardware..."})
     env = HardwareEnv.detect()
     perf = PerfTracker(interval=PERF_STATS_INTERVAL)
 
+    gpu_msg = f"{env.gpu_name} ({env.backend})" if env.gpu_name else env.backend
+    emit({"event": "progress", "stage": "init", "message": f"Hardware: {gpu_msg}"})
+
     try:
+        emit({"event": "progress", "stage": "model", "message": f"Loading {model_name} model ({env.export_format} format)..."})
         model, model_format = env.load_optimized(model_name, use_optimized=use_optimized)
         perf.model_load_ms = env.load_ms
         perf.export_ms = env.export_ms
 
+        if env.export_ms > 0:
+            emit({"event": "progress", "stage": "model", "message": f"Model optimized in {env.export_ms:.0f}ms"})
+
         ready_event = {
             "event": "ready",
             "model": f"yolo2026{model_size[0]}",
@@ -268,18 +276,19 @@ def main():
         emit({"event": "error", "message": f"Failed to load model: {e}", "retriable": False})
         sys.exit(1)
 
-    # Graceful shutdown
-    running = True
+    # Graceful shutdown — exit immediately with code 0.
+    # The stdin read loop blocks, so setting a flag doesn't work;
+    # we must exit in the signal handler to avoid being killed (code null).
     def handle_signal(signum, frame):
-        nonlocal running
-        running = False
+        sig_name = "SIGTERM" if signum == signal.SIGTERM else "SIGINT"
+        log(f"Received {sig_name}, shutting down gracefully")
+        perf.emit_final()
+        sys.exit(0)
     signal.signal(signal.SIGTERM, handle_signal)
     signal.signal(signal.SIGINT, handle_signal)
 
     # Main loop: read frames from stdin, output detections to stdout
     for line in sys.stdin:
-        if not running:
-            break
 
         line = line.strip()
         if not line:

skills/lib/env_config.py

@@ -133,31 +133,79 @@ def detect() -> "HardwareEnv":
         return env
 
     def _try_cuda(self) -> bool:
-        """Detect NVIDIA GPU via nvidia-smi and torch."""
-        if not shutil.which("nvidia-smi"):
-            return False
+        """Detect NVIDIA GPU via nvidia-smi (with Windows path search) and WMI fallback."""
+        nvidia_smi = shutil.which("nvidia-smi")
+
+        # Windows: check well-known paths if not on PATH
+        if not nvidia_smi and platform.system() == "Windows":
+            for candidate in [
+                Path(os.environ.get("PROGRAMFILES", r"C:\Program Files"))
+                / "NVIDIA Corporation" / "NVSMI" / "nvidia-smi.exe",
+                Path(os.environ.get("WINDIR", r"C:\Windows"))
+                / "System32" / "nvidia-smi.exe",
+            ]:
+                if candidate.is_file():
+                    nvidia_smi = str(candidate)
+                    _log(f"Found nvidia-smi at {nvidia_smi}")
+                    break
+
+        if nvidia_smi:
+            try:
+                result = subprocess.run(
+                    [nvidia_smi, "--query-gpu=name,memory.total,driver_version",
+                     "--format=csv,noheader,nounits"],
+                    capture_output=True, text=True, timeout=10,
+                )
+                if result.returncode == 0:
+                    line = result.stdout.strip().split("\n")[0]
+                    parts = [p.strip() for p in line.split(",")]
+                    if len(parts) >= 3:
+                        self.backend = "cuda"
+                        self.device = "cuda"
+                        self.gpu_name = parts[0]
+                        self.gpu_memory_mb = int(float(parts[1]))
+                        self.driver_version = parts[2]
+                        self.detection_details["nvidia_smi"] = line
+                        _log(f"NVIDIA GPU: {self.gpu_name} ({self.gpu_memory_mb}MB, driver {self.driver_version})")
+                        return True
+            except (subprocess.TimeoutExpired, FileNotFoundError, ValueError) as e:
+                _log(f"nvidia-smi probe failed: {e}")
+
+        # Windows WMI fallback: detect NVIDIA GPU even without nvidia-smi on PATH
+        if platform.system() == "Windows":
+            return self._try_cuda_wmi()
+
+        return False
+
+    def _try_cuda_wmi(self) -> bool:
+        """Windows-only: detect NVIDIA GPU via WMI (wmic)."""
         try:
             result = subprocess.run(
-                ["nvidia-smi", "--query-gpu=name,memory.total,driver_version",
-                 "--format=csv,noheader,nounits"],
+                ["wmic", "path", "win32_VideoController", "get",
+                 "Name,AdapterRAM,DriverVersion", "/format:csv"],
                 capture_output=True, text=True, timeout=10,
             )
             if result.returncode != 0:
                 return False
 
-            line = result.stdout.strip().split("\n")[0]
-            parts = [p.strip() for p in line.split(",")]
-            if len(parts) >= 3:
-                self.backend = "cuda"
-                self.device = "cuda"
-                self.gpu_name = parts[0]
-                self.gpu_memory_mb = int(float(parts[1]))
-                self.driver_version = parts[2]
-                self.detection_details["nvidia_smi"] = line
-                _log(f"NVIDIA GPU: {self.gpu_name} ({self.gpu_memory_mb}MB, driver {self.driver_version})")
-                return True
+            for line in result.stdout.strip().split("\n"):
+                if "NVIDIA" in line.upper():
+                    parts = [p.strip() for p in line.split(",")]
+                    # CSV format: Node,AdapterRAM,DriverVersion,Name
+                    if len(parts) >= 4:
+                        self.backend = "cuda"
+                        self.device = "cuda"
+                        self.gpu_name = parts[3]
+                        try:
+                            self.gpu_memory_mb = int(int(parts[1]) / (1024 * 1024))
+                        except (ValueError, IndexError):
+                            pass
+                        self.driver_version = parts[2] if len(parts) > 2 else ""
+                        self.detection_details["wmi"] = line
+                        _log(f"NVIDIA GPU (WMI): {self.gpu_name} ({self.gpu_memory_mb}MB)")
+                        return True
         except (subprocess.TimeoutExpired, FileNotFoundError, ValueError) as e:
-            _log(f"nvidia-smi probe failed: {e}")
+            _log(f"WMI probe failed: {e}")
         return False
 
     def _try_rocm(self) -> bool: