[SOLVED] Problem with tvmc run

Troubleshooting
#1

I compiled a tflite model using tvmc and when I run the model I get the following error.

Error: the input tensor 'data' is not in the graph. Expected inputs: 'dict_keys(['input'])'

Could someone please guide me to how to debug this problem.

EDIT:

Target is Raspberry Pi4

I compiled the facedetection tflite model from media pipe.

tvmc compile --target "llvm -mtriple=arm-linux-gnueabihf" --output "model.tar" models/face_detection_full_range.tflite

Then preprocessed the input image into a numpy array as described in the documentation and then used that for inference.

tvmc run --inputs face.npz --outputs preds.npz model.tar

1 Like
#2

Hi @abhilb, can you provide a bit more detail on the command lines you ran, and perhaps a bit more about the model you are using?

#4

Based on the message, is that possible that your model is expecting an input called input, and you are exporting your .npz input file with an input called data?

In the sample input pre-processing script, this is set in the very last line:

np.savez("imagenet_cat", data=img_data)
                         ^^^^

Can you try replacing that data= with input=, so that it matches with your model?

I can try to debug this further if needed. Let me know.

4 Likes
#5

@leandron That was it. I never thought about it. I thought I was doing something wrong in the compilation step. Thanks a lot.

1 Like