aurotripathy · March 10, 2026 19:06 · aurotripathy · Mar 10, 2026
diff --git a/sample-pt-load-convert-to-onnx-infer.py b/sample-pt-load-convert-to-onnx-infer.py
 import onnxruntime
 import numpy as np

 from ultralytics import YOLO
 # Load a pretrained YOLOv8 model                                                                                                                           
 model = YOLO('yolov8n.pt')
 # Export to ONNX format                                                                                                                                    
 model.export(format='onnx')


 # 1. Load the ONNX model into an Inference Session                                                                                                         
 # Specify execution providers, e.g., for GPU/CPU                                                                                                           
 session = onnxruntime.InferenceSession("yolov8n.onnx", providers=['CUDAExecutionProvider', 'CPUExecutionProvider'])

 # 2. Prepare the input data (example with a dummy input)                                                                                                   
 # The input name should match the name defined in the model (check with Netron)                                                                            
 input_name = session.get_inputs()[0].name
 # Create a dummy input tensor matching the model's expected shape and type                                                                                 
 # (e.g., a float32 tensor of shape [1, 3, 224, 224] for an image model)                                                                                    
 input_data = np.random.rand(1, 3, 640, 640).astype(np.float32)
 inputs = {input_name: input_data}

 # 3. Run the model                                                                                                                                         
 # Pass an empty list for the first argument to get all outputs                                                                                             
 outputs = session.run(None, inputs)

 # 4. Post-process the output                                                                                                                               
 # The result is a list of outputs; access the first one here                                                                                               
 model_output = outputs[0]
 print(model_output)
	import onnxruntime
	import numpy as np

	from ultralytics import YOLO
	# Load a pretrained YOLOv8 model
	model = YOLO('yolov8n.pt')
	# Export to ONNX format
	model.export(format='onnx')


	# 1. Load the ONNX model into an Inference Session
	# Specify execution providers, e.g., for GPU/CPU
	session = onnxruntime.InferenceSession("yolov8n.onnx", providers=['CUDAExecutionProvider', 'CPUExecutionProvider'])

	# 2. Prepare the input data (example with a dummy input)
	# The input name should match the name defined in the model (check with Netron)
	input_name = session.get_inputs()[0].name
	# Create a dummy input tensor matching the model's expected shape and type
	# (e.g., a float32 tensor of shape [1, 3, 224, 224] for an image model)
	input_data = np.random.rand(1, 3, 640, 640).astype(np.float32)
	inputs = {input_name: input_data}

	# 3. Run the model
	# Pass an empty list for the first argument to get all outputs
	outputs = session.run(None, inputs)

	# 4. Post-process the output
	# The result is a list of outputs; access the first one here
	model_output = outputs[0]
	print(model_output)
No results found