How to use yolo8x

[AlexanderKozhevin]

I converted it to tflite. But it gives an error like this

Cannot copy to a TensorFlowLite tensor (serving_default_images:0) with 4915200 bytes from a Java Buffer with 23040 bytes.

[AlexanderKozhevin]

I checked the model from example and it has the following parameters

Input Details: [{'name': 'inputs_0', 'index': 0, 'shape': array([  1, 224, 224,   3], dtype=int32), 'shape_signature': array([  1, 224, 224,   3], dtype=int32), 'dtype': <class 'numpy.float32'>, 'quantization': (0.0, 0), 'quantization_parameters': {'scales': array([], dtype=float32), 'zero_points': array([], dtype=int32), 'quantized_dimension': 0}, 'sparsity_parameters': {}}]

Output Details: [{'name': 'Identity', 'index': 537, 'shape': array([   1,   84, 1029], dtype=int32), 'shape_signature': array([   1,   84, 1029], dtype=int32), 'dtype': <class 'numpy.float32'>, 'quantization': (0.0, 0), 'quantization_parameters': {'scales': array([], dtype=float32), 'zero_points': array([], dtype=int32), 'quantized_dimension': 0}, 'sparsity_parameters': {}}]

But my model has different

Input Details: [{'name': 'serving_default_images:0', 'index': 0, 'shape': array([  1,   3, 640, 640], dtype=int32), 'shape_signature': array([  1,   3, 640, 640], dtype=int32), 'dtype': <class 'numpy.float32'>, 'quantization': (0.0, 0), 'quantization_parameters': {'scales': array([], dtype=float32), 'zero_points': array([], dtype=int32), 'quantized_dimension': 0}, 'sparsity_parameters': {}}]

Output Details: [{'name': 'PartitionedCall:0', 'index': 572, 'shape': array([   1,   84, 8400], dtype=int32), 'shape_signature': array([   1,   84, 8400], dtype=int32), 'dtype': <class 'numpy.float32'>, 'quantization': (0.0, 0), 'quantization_parameters': {'scales': array([], dtype=float32), 'zero_points': array([], dtype=int32), 'quantized_dimension': 0}, 'sparsity_parameters': {}}]

All yolo models has image size 640px. How could it be 224?

[habbas11]

I believe you could change the input size while converting your model from .pt to .tflite
model.export(format='tflite', imgsz=224)

[guiiyanxD]

I checked the model from example and it has the following parameters

Input Details: [{'name': 'inputs_0', 'index': 0, 'shape': array([  1, 224, 224,   3], dtype=int32), 'shape_signature': array([  1, 224, 224,   3], dtype=int32), 'dtype': <class 'numpy.float32'>, 'quantization': (0.0, 0), 'quantization_parameters': {'scales': array([], dtype=float32), 'zero_points': array([], dtype=int32), 'quantized_dimension': 0}, 'sparsity_parameters': {}}]

Output Details: [{'name': 'Identity', 'index': 537, 'shape': array([   1,   84, 1029], dtype=int32), 'shape_signature': array([   1,   84, 1029], dtype=int32), 'dtype': <class 'numpy.float32'>, 'quantization': (0.0, 0), 'quantization_parameters': {'scales': array([], dtype=float32), 'zero_points': array([], dtype=int32), 'quantized_dimension': 0}, 'sparsity_parameters': {}}]

But my model has different

Input Details: [{'name': 'serving_default_images:0', 'index': 0, 'shape': array([  1,   3, 640, 640], dtype=int32), 'shape_signature': array([  1,   3, 640, 640], dtype=int32), 'dtype': <class 'numpy.float32'>, 'quantization': (0.0, 0), 'quantization_parameters': {'scales': array([], dtype=float32), 'zero_points': array([], dtype=int32), 'quantized_dimension': 0}, 'sparsity_parameters': {}}]

Output Details: [{'name': 'PartitionedCall:0', 'index': 572, 'shape': array([   1,   84, 8400], dtype=int32), 'shape_signature': array([   1,   84, 8400], dtype=int32), 'dtype': <class 'numpy.float32'>, 'quantization': (0.0, 0), 'quantization_parameters': {'scales': array([], dtype=float32), 'zero_points': array([], dtype=int32), 'quantized_dimension': 0}, 'sparsity_parameters': {}}]

All yolo models has image size 640px. How could it be 224?

Have you achieved to fit the inputs size? I'm facing same problem right now. When exporting my yolov8n-seg to tflite I specified imgsz=640 but when running the model on frame it throws the same error. How did you solve this? Please. Thank you for your reply in advanced

[pwilliamspeniel]

@guiiyanxD and @AlexanderKozhevin. How were you guys able to resolve the input tensor shape, because the model shape for the tensor here is [(tensor: float32[1,640,640,3]) while i have tensor: float32[1,3,640,640]. I don't know why it is so.