ints8_quantization, as executorch is failing

README.md

@@ -18,7 +18,7 @@ Full-stack traffic analytics dashboard. YOLO + ByteTrack backend processes uploa
 
 ## Stack
 
-- **Backend**: FastAPI, Ultralytics YOLO (ExecuTorch), ByteTrack, OpenCV
+- **Backend**: FastAPI, Ultralytics YOLO (OpenVINO), ByteTrack, OpenCV
 - **Frontend**: Vanilla HTML/JS, TailwindCSS CDN, Chart.js
 - **Infra**: Docker, CPU-only inference
 

backend/model.py

@@ -7,10 +7,10 @@ load_dotenv()
 
 MODEL_DIR = Path(__file__).parent / "weights"
 PT_PATH = MODEL_DIR / "best.pt"
-ET_DIR = MODEL_DIR / "yolo26n_executorch_model"
+OV_DIR = MODEL_DIR / "best_int8_openvino_model"
 
 
-def ensure_executorch():
+def ensure_openvino():
     MODEL_DIR.mkdir(exist_ok=True)
 
     if not PT_PATH.exists():
@@ -21,11 +21,11 @@ def ensure_executorch():
             f'https://huggingface.co/Perception365/VehicleNet-Y26s/resolve/main/weights/best.pt'
         )
 
-    # https://docs.ultralytics.com/integrations/executorch/
-    if not ET_DIR.exists():
-        YOLO(str(PT_PATH)).export(format="executorch", device='cpu')
+    # https://docs.ultralytics.com/modes/export/#arguments
+    if not OV_DIR.exists():
+        YOLO(str(PT_PATH)).export(format="openvino", dynamic=True, int8=True)
 
 
 def load_model():
-    ensure_executorch()
-    return YOLO(str(ET_DIR), task="detect")
+    ensure_openvino()
+    return YOLO(str(OV_DIR), task="detect")

frontend/vehicles.html

@@ -421,7 +421,7 @@
                             <ul class="text-xs space-y-3 pl-1">
                                 <li class="flex items-start gap-3"><i
                                         class="fa-solid fa-circle text-[5px] mt-1.5 text-slate-400"></i>
-                                    <span>ExecuTorch Performance Optimization</span>
+                                    <span>OpenVINO Performance Optimization</span>
                                 </li>
                                 <li class="flex items-start gap-3"><i
                                         class="fa-solid fa-circle text-[5px] mt-1.5 text-slate-400"></i>
@@ -888,7 +888,7 @@
                     <a href="https://huggingface.co/Perception365/VehicleNet-Y26s" target="_blank" class="text-sm font-bold text-white mono-font hover:text-slate-300 transition underline underline-offset-4 decoration-slate-700">Perception365/VehicleNet-Y26s</a>
                 </div>` +
                 detailRow('task', 'detect') +
-                detailRow('format', 'ExecuTorch') +
+                detailRow('format', 'OpenVINO') +
                 detailRow('tracker', 'ByteTrack');
 
             populateInferPanel(c);

requirements.txt

@@ -4,10 +4,13 @@ python-multipart==0.0.6
 opencv-python-headless==4.8.1.78
 matplotlib>=3.7.0
 ultralytics>=8.3.0
-numpy==1.24.3
+numpy>=1.23.0,<2.0.0
 python-dotenv==1.0.0
 websockets==12.0
-executorch
+onnxruntime
+openvino>=2024.0.0
+onnx>=1.12.0
+onnxslim>=0.1.71
 lap>=0.5.12
 torch==2.1.0+cpu
 torchvision==0.16.0+cpu