Could Not Work with Multi-Input Tflite Model


I was trying to upload my trained tflite model which has multiple inputs. But I received error, “Expected a single input, but found 3”. I was wondering, if Edge Impulse supports multi-input models. If so, please let me know any possible pointer to take a look into.



Hi @hnahian1 We currently don’t.

One way around it could be to rewrite your saved model graph to create a single input tensor, and then slice that into three tensors (tensor slicing) and then run the rest of your network. I’ve done stuff like this with onnx graph surgeon in the past, and you can theoretically do it w/o retraining model.

What kind of data are you processing that you have 3 input tensors?