Hello, I have a NN with layout B, C, H, W and shape 1, 300, 300, 3 ([open_model_zoo here](https://github.com/openvinotoolkit/open_model_zoo/tree/master/models/intel/vehicle-license-plate-detection-barrier-0106)) have OAK-1 lite camera, whose preview of shape 300, 300 is linked do this NN. I constantly get warning: `[18443010C1AA670F00] [1.1.1] [4.653] [NeuralNetwork(2)] [warning] Input image (300x300) does not match NN (3x300)` What can I do to make it work? Cheers

Hi @"Rafek"#3046 , Likely you mixed CHW/HWC (planar vs interleaved). You can try: - Toggling `colorCam.setInterleaved(False)` (or True) - Specify channels when converting model, [docs here](https://docs.luxonis.com/software/ai-inference/conversion/#Conversion-Advanced%20Settings-Model%20Optimizer%20Flags-Model%20Layout)

Wrong NN inpur shape

Rafek

Hello,

I have a NN with layout B, C, H, W and shape 1, 300, 300, 3 (open_model_zoo here) have OAK-1 lite camera, whose preview of shape 300, 300 is linked do this NN. I constantly get warning:

[18443010C1AA670F00] [1.1.1] [4.653] [NeuralNetwork(2)] [warning] Input image (300x300) does not match NN (3x300)

What can I do to make it work?

Cheers

erik

Hi @Rafek ,
Likely you mixed CHW/HWC (planar vs interleaved). You can try:

Toggling colorCam.setInterleaved(False) (or True)
Specify channels when converting model, docs here

Rafek

I tried setting cam.setInterleaved(True) (instead of False) and I got this warning:

[18443010C1AA670F00] [1.1.4] [4.745] [NeuralNetwork(2)] [error] Input tensor 'Placeholder' (0) exceeds available data range. Data size (196608B), tensor offset (0), size (270000B) - skipping inference
[18443010C1AA670F00] [1.1.4] [4.946] [NeuralNetwork(2)] [warning] Input image (256x256) does not match NN (3x300)
[18443010C1AA670F00] [1.1.4] [4.946] [NeuralNetwork(2)] [warning] Input image is interleaved (HWC), NN specifies planar (CHW) data order

The nn is the one I mentioned above. Is there anything I can do to make it work?

jakaskerl

Rafek
Interleaved option should be off.

Rafek Input image (256x256)

This indicates a wrong preview size is set. Change it back to 300x300. As Erik said, specify the model channels when converting. You can also (though not completely sure) change to an older openVino version conversion, I think around 2021.4 the default channel layout was swapped.

Thanks,
Jaka