Open ron-deepkeep opened 1 year ago
My first thought is about quantization/dequantization which might cause some small data loss. But it needs more efforts to compare the results of the same op in different formats (TensorFlow/ONNX) so that we might find the root cause for the similar results.
Ask a Question
Question
when I convert my model from tflite to onnx and try out the same image on both the inference is similar, but not exactly identical would you might have any idea why, and if so is there a remedy?