Mini CNN has a bug?

[SrMarkLau]

Hi friend,

I ran the code of mini ccn, but python returns:

`ValueError: Exception encountered when calling layer "daub_wave_layer2d" (type DaubWaveLayer2D).

in user code:

File "C:\Users\marco\AppData\Local\Programs\Python\Python39\lib\site-packages\wavetf\_base_wavelets.py", line 109, in call  *
    return self.kernel_function(batch)
File "C:\Users\marco\AppData\Local\Programs\Python\Python39\lib\site-packages\wavetf\_daubachies_conv.py", line 216, in kernel_function  *
    s1 = tf.reshape(s1, [self.bs*self.cn*self.ox, -1, 1])

ValueError: Invalid value in tensor used for shape: -672

Call arguments received: • batch=tf.Tensor(shape=(None, 224, 224, 3), dtype=float32)`

The images are RGB:

img_size= 224 channels = 3 input_shape = (img_size, img_size, channels)

[fversaci]

Hi @SrMarkLau, which version of Tensorflow are you using?

[thegodone]

same for me tf 2.4.0

with 128x128x3 as input shape

ValueError: Invalid value in tensor used for shape: -384

same with tf2.2

i just run this command in Jupyter wavelet_cnn(input_shape=(128,128,3))

[fversaci]

Hi, can you produce a minimal, non-working example that I could run and test?

[thegodone]

my error

ValueError Traceback (most recent call last)

in ----> 1 wavelet_cnn(input_shape=(128,128,3)) in wavelet_cnn(input_shape, ks, baselev, wavelet, wave_kern, hsv, convrep, num_classes) 22 wave0 = inputs 23 # compute 4 level of wavelet ---> 24 wave1 = WaveTFFactory.build(wave_kern)(wave0) 25 # compute new wavelet features from LL componenents 26 wave2 = WaveTFFactory.build(wave_kern)(wave1[:,:,:,:chans]) /usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/engine/base_layer.py in __call__(self, *args, **kwargs) 920 not base_layer_utils.is_in_eager_or_tf_function()): 921 with auto_control_deps.AutomaticControlDependencies() as acd: --> 922 outputs = call_fn(cast_inputs, *args, **kwargs) 923 # Wrap Tensors in `outputs` in `tf.identity` to avoid 924 # circular dependencies. /usr/local/lib/python3.6/dist-packages/tensorflow/python/autograph/impl/api.py in wrapper(*args, **kwargs) 263 except Exception as e: # pylint:disable=broad-except 264 if hasattr(e, 'ag_error_metadata'): --> 265 raise e.ag_error_metadata.to_exception(e) 266 else: 267 raise ValueError: in user code: /usr/local/lib/python3.6/dist-packages/wavetf/_base_wavelets.py:109 call * return self.kernel_function(batch) /usr/local/lib/python3.6/dist-packages/wavetf/_daubachies_conv.py:216 kernel_function * s1 = tf.reshape(s1, [self.bs*self.cn*self.ox, -1, 1]) /usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/array_ops.py:193 reshape ** result = gen_array_ops.reshape(tensor, shape, name) /usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/gen_array_ops.py:8087 reshape "Reshape", tensor=tensor, shape=shape, name=name) /usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/op_def_library.py:744 _apply_op_helper attrs=attr_protos, op_def=op_def) /usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/func_graph.py:595 _create_op_internal compute_device) /usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/ops.py:3327 _create_op_internal op_def=op_def) /usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/ops.py:1817 __init__ control_input_ops, op_def) /usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/ops.py:1657 _create_c_op raise ValueError(str(e)) ValueError: Invalid value in tensor used for shape: -384

[thegodone]

I clone your repository master branch and next pip install .

[fversaci]

Hi, I just pushed a commit that seems to fix this issue. Could you check if that works for you as well?

[thegodone]

fix for me thanks a lot

[fversaci]

All right then :+1: Thank you for pointing out the issue!

[thegodone]

Welcome, small question: do you know how to reproduce this code with your keras module ? https://github.com/mxbastidasr/DAWN_WACV2020 (Wavelet CNN part)

[fversaci]

Hi, I don't have time to write the network in detail, but keep in mind that, being wave the output of a wavelet layer working on 3 channels (e.g., RGB), its LL component is given by wave[..., :3], while the LH, HL and HH components are wave[..., 3:].

All the other layers in your network look like standard keras ones. Try to have a look at mini_cnn.py code for an example of applying a wavelet layer on the LL component of the output of a previous wavelet layer.