add export to FP16

add inference with possibility to have different

inference/start_inference.py

@@ -0,0 +1,20 @@
+from onnx_engine import OnnxEngine
+from tensorrt_engine import TensorRTEngine
+from inference import Inference
+
+if __name__ == "__main__":
+    # Inference(OnnxEngine('azaion-2025-03-10.onnx', batch_size=4),
+    #            confidence_threshold=0.5, iou_threshold=0.3).process('ForAI_test.mp4')
+    # detection for the first 200sec of video:
+    # onnxInference: 81 sec, 6.3Gb VRAM
+    # tensorrt: 54 sec, 3.7Gb VRAM
+
+    # Inference(TensorRTEngine('azaion-2025-03-10_int8.engine', batch_size=16),
+    #          confidence_threshold=0.5, iou_threshold=0.3).process('ForAI_test.mp4')
+    # INT8 for 200sec: 54 sec 3.7Gb
+
+    # Inference(TensorRTEngine('azaion-2025-03-10_batch8.engine', batch_size=8),
+    #           confidence_threshold=0.5, iou_threshold=0.3).process('ForAI_test.mp4')
+
+    Inference(TensorRTEngine('azaion-2025-03-10-half_batch4.engine', batch_size=4),
+               confidence_threshold=0.5, iou_threshold=0.3).process('ForAI_test.mp4')