jihongju/keras-resnet3d

Implementations of ResNets for volumetric data, including a vanilla resnet in 3D.

keras-resnet3d

Resnet 3D

A vanilla 3D extention to raghakot/keras-resnet

VoxResNet (TODO)

A keras re-implementation of VoxResNet (Hao Chen et.al) for volumetric image segmention. (Non-official)

keras-voxresnet enables volumetric image classification with keras and tensorflow/theano.

Installation

Dependencies

keras, tensorflow/theano and their corresponding dependencies.

Install with pip

$ pip install git+https://github.com/JihongJu/keras-resnet3d.git

Build from source

$ git clone https://github.com/JihongJu/keras-resnet3d.git
$ cd keras-resnet3d
$ python setup.py build

Usage

from resnet3d import Resnet3DBuilder
model = Resnet3DBuilder.build_resnet_50((96, 96, 96, 1), 20)
model.compile(optimizer='adam',
              loss='categorical_crossentropy',
              metrics=['accuracy'])
model.fit(X_train, y_train, batch_size=32)

More details see example:

• A lung-cancer-detector

keras-resnet3d's Issues

Pretrained models of ResNet 3D for Keras

Hello and thank you for your contribution,

Can you tell us if there are pretrained models of ResNet 3D for Tensorflow/Keras (.hdf5 or .h5 model formats) ?
I found that there are pretrained models of this architecture for Pytorch here https://github.com/kenshohara/3D-ResNets-PyTorch

Import Format - Batch Normalization

It seems that there is an issue when importing 'BatchNormalization' from 'keras.layers.normalization' in the script [resnet3d.py]

The import should be done in a different way:

from tensorflow.keras.layers import BatchNormalization

as opposed to

from keras.layers.normalization import BatchNormalization

Maybe the issue is due to some underlying problem.

I am installing this package with the command below in google colab

! pip install git+https://github.com/JihongJu/keras-resnet3d.git
and creating a modle like so:

from resnet3d import Resnet3DBuilder
model = Resnet3DBuilder.build_resnet_50((96, 96, 96, 1), 20)

Pretrained weight for transfer learning

Dear @JihongJu

Where should i find the pretrained weight of this model in term of transfer learning?

Best Regards

Input shape

I want to take input shape like [32,4,256,128,3] as Batch_size x Time x H x W x C, what is the order of dim1,dim2,dim3,... at the input shape

Problem taken variable input shape (None,None,None,num_ch)

Hi Jihong,

I really like your code. I'm training a patch-based network classifier and now I want to test it on a whole 3D volume. I know that in Keras, if you specify input shape as (None,None,None,num_ch), then you can test on a whole image and it will evaluate patch-wise. This is much faster than sampling manually since it uses the GPU.

However, when I try that with you repo I get the following error:

model = Resnet3DBuilder.build_resnet_18((None, None, None, 2), 1)
File "/tensorflow/lib/python2.7/site-packages/resnet3d/resnet3d.py", line 272, in build_resnet_18
[2, 2, 2, 2], reg_factor=reg_factor)
File "/tensorflow/lib/python2.7/site-packages/resnet3d/resnet3d.py", line 242, in build
)(block)
File "/tensorflow/lib/python2.7/site-packages/resnet3d/resnet3d.py", line 112, in f
)(input)
File "/tensorflow/lib/python2.7/site-packages/resnet3d/resnet3d.py", line 139, in f
return _shortcut3d(input, residual)
File "/tensorflow/lib/python2.7/site-packages/resnet3d/resnet3d.py", line 80, in _shortcut3d
// residual._keras_shape[DIM1_AXIS]
TypeError: unsupported operand type(s) for //: 'NoneType' and 'NoneType'

I'm wondering if you've thought about a way to fix this. It would be immensely useful.

Thanks!

Memory issue

I keep receiving memory allocation error. which make sense regarding my GPU size (12 Gb). Do you guys get the same error? (input size is [64, 64, 64], batch size: 1 !)

Problem on test set

Dear All,
I try to install on Ubuntu 18 using conda with python3.6:
here the error for sh test.sh

`lugins: xdist-2.1.0, cov-2.10.1, forked-1.3.0
collected 7 items                                                                                                                                                             

tests/test_resnet3d.py FFFFF.F                                                                                                                                          [100%]

================================================================================== FAILURES ===================================================================================
______________________________________________________________________________ test_resnet3d_18 _______________________________________________________________________________

resnet3d_test = <function resnet3d_test.<locals>.f at 0x7f3c1abbd048>

    def test_resnet3d_18(resnet3d_test):
        """Test 18."""
        K.set_image_data_format('channels_last')
>       model = Resnet3DBuilder.build_resnet_18((224, 224, 224, 1), 2)

tests/test_resnet3d.py:20: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
resnet3d/resnet3d.py:273: in build_resnet_18
    [2, 2, 2, 2], reg_factor=reg_factor)
resnet3d/resnet3d.py:243: in build
    )(block)
resnet3d/resnet3d.py:113: in f
    )(input)
resnet3d/resnet3d.py:140: in f
    return _shortcut3d(input, residual)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

input = <tf.Tensor 'max_pooling3d/MaxPool3D:0' shape=(None, 56, 56, 56, 64) dtype=float32>
residual = <tf.Tensor 'conv3d_2/BiasAdd:0' shape=(None, 56, 56, 56, 64) dtype=float32>

    def _shortcut3d(input, residual):
        """3D shortcut to match input and residual and merges them with "sum"."""
>       stride_dim1 = ceil(input._keras_shape[DIM1_AXIS] \
            / residual._keras_shape[DIM1_AXIS])
E       AttributeError: 'Tensor' object has no attribute '_keras_shape'

resnet3d/resnet3d.py:80: AttributeError
---------------------------------------------------------------------------- Captured stderr call -----------------------------------------------------------------------------
2020-10-15 17:33:12.626916: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcuda.so.1
2020-10-15 17:33:12.672638: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:982] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-10-15 17:33:12.672836: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1716] Found device 0 with properties: 
pciBusID: 0000:01:00.0 name: Quadro RTX 4000 computeCapability: 7.5
coreClock: 1.545GHz coreCount: 36 deviceMemorySize: 7.79GiB deviceMemoryBandwidth: 387.49GiB/s
2020-10-15 17:33:12.672879: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library 'libcudart.so.10.1'; dlerror: libcudart.so.10.1: cannot open shared object file: No such file or directory
2020-10-15 17:33:12.672911: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library 'libcublas.so.10'; dlerror: libcublas.so.10: cannot open shared object file: No such file or directory
2020-10-15 17:33:12.672942: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library 'libcufft.so.10'; dlerror: libcufft.so.10: cannot open shared object file: No such file or directory
2020-10-15 17:33:12.672972: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library 'libcurand.so.10'; dlerror: libcurand.so.10: cannot open shared object file: No such file or directory
2020-10-15 17:33:12.673001: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library 'libcusolver.so.10'; dlerror: libcusolver.so.10: cannot open shared object file: No such file or directory
2020-10-15 17:33:12.673029: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library 'libcusparse.so.10'; dlerror: libcusparse.so.10: cannot open shared object file: No such file or directory
2020-10-15 17:33:12.673059: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library 'libcudnn.so.7'; dlerror: libcudnn.so.7: cannot open shared object file: No such file or directory
2020-10-15 17:33:12.673065: W tensorflow/core/common_runtime/gpu/gpu_device.cc:1753] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU. Follow the guide at https://www.tensorflow.org/install/gpu for how to download and setup the required libraries for your platform.
Skipping registering GPU devices...
2020-10-15 17:33:12.673215: I tensorflow/core/platform/cpu_feature_guard.cc:142] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN)to use the following CPU instructions in performance-critical operations:  AVX2 FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
2020-10-15 17:33:12.697604: I tensorflow/core/platform/profile_utils/cpu_utils.cc:104] CPU Frequency: 3600000000 Hz
2020-10-15 17:33:12.698881: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x55f1090d5040 initialized for platform Host (this does not guarantee that XLA will be used). Devices:
2020-10-15 17:33:12.698894: I tensorflow/compiler/xla/service/service.cc:176]   StreamExecutor device (0): Host, Default Version
2020-10-15 17:33:12.699862: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1257] Device interconnect StreamExecutor with strength 1 edge matrix:
2020-10-15 17:33:12.699871: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1263]      
______________________________________________________________________________ test_resnet3d_34 _______________________________________________________________________________

resnet3d_test = <function resnet3d_test.<locals>.f at 0x7f3c184f0ae8>

    def test_resnet3d_34(resnet3d_test):
        """Test 34."""
        K.set_image_data_format('channels_last')
>       model = Resnet3DBuilder.build_resnet_34((224, 224, 224, 1), 2)

tests/test_resnet3d.py:30: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
resnet3d/resnet3d.py:279: in build_resnet_34
    [3, 4, 6, 3], reg_factor=reg_factor)
resnet3d/resnet3d.py:243: in build
    )(block)
resnet3d/resnet3d.py:113: in f
    )(input)
resnet3d/resnet3d.py:140: in f
    return _shortcut3d(input, residual)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

input = <tf.Tensor 'max_pooling3d_1/MaxPool3D:0' shape=(None, 56, 56, 56, 64) dtype=float32>
residual = <tf.Tensor 'conv3d_5/BiasAdd:0' shape=(None, 56, 56, 56, 64) dtype=float32>

    def _shortcut3d(input, residual):
        """3D shortcut to match input and residual and merges them with "sum"."""
>       stride_dim1 = ceil(input._keras_shape[DIM1_AXIS] \
            / residual._keras_shape[DIM1_AXIS])
E       AttributeError: 'Tensor' object has no attribute '_keras_shape'

resnet3d/resnet3d.py:80: AttributeError
______________________________________________________________________________ test_resnet3d_50 _______________________________________________________________________________

resnet3d_test = <function resnet3d_test.<locals>.f at 0x7f3c1aa8cb70>

    def test_resnet3d_50(resnet3d_test):
        """Test 50."""
        K.set_image_data_format('channels_last')
>       model = Resnet3DBuilder.build_resnet_50((224, 224, 224, 1), 1, 1e-2)

tests/test_resnet3d.py:40: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
resnet3d/resnet3d.py:285: in build_resnet_50
    [3, 4, 6, 3], reg_factor=reg_factor)
resnet3d/resnet3d.py:243: in build
    )(block)
resnet3d/resnet3d.py:113: in f
    )(input)
resnet3d/resnet3d.py:169: in f
    return _shortcut3d(input, residual)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

input = <tf.Tensor 'max_pooling3d_2/MaxPool3D:0' shape=(None, 56, 56, 56, 64) dtype=float32>
residual = <tf.Tensor 'conv3d_9/BiasAdd:0' shape=(None, 56, 56, 56, 256) dtype=float32>

    def _shortcut3d(input, residual):
        """3D shortcut to match input and residual and merges them with "sum"."""
>       stride_dim1 = ceil(input._keras_shape[DIM1_AXIS] \
            / residual._keras_shape[DIM1_AXIS])
E       AttributeError: 'Tensor' object has no attribute '_keras_shape'

resnet3d/resnet3d.py:80: AttributeError
______________________________________________________________________________ test_resnet3d_101 ______________________________________________________________________________

resnet3d_test = <function resnet3d_test.<locals>.f at 0x7f3c1abbd0d0>

    def test_resnet3d_101(resnet3d_test):
        """Test 101."""
        K.set_image_data_format('channels_last')
>       model = Resnet3DBuilder.build_resnet_101((224, 224, 224, 1), 2)

tests/test_resnet3d.py:50: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
resnet3d/resnet3d.py:291: in build_resnet_101
    [3, 4, 23, 3], reg_factor=reg_factor)
resnet3d/resnet3d.py:243: in build
    )(block)
resnet3d/resnet3d.py:113: in f
    )(input)
resnet3d/resnet3d.py:169: in f
    return _shortcut3d(input, residual)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

input = <tf.Tensor 'max_pooling3d_3/MaxPool3D:0' shape=(None, 56, 56, 56, 64) dtype=float32>
residual = <tf.Tensor 'conv3d_13/BiasAdd:0' shape=(None, 56, 56, 56, 256) dtype=float32>

    def _shortcut3d(input, residual):
        """3D shortcut to match input and residual and merges them with "sum"."""
>       stride_dim1 = ceil(input._keras_shape[DIM1_AXIS] \
            / residual._keras_shape[DIM1_AXIS])
E       AttributeError: 'Tensor' object has no attribute '_keras_shape'

resnet3d/resnet3d.py:80: AttributeError
______________________________________________________________________________ test_resnet3d_152 ______________________________________________________________________________

resnet3d_test = <function resnet3d_test.<locals>.f at 0x7f3c184b2f28>

    def test_resnet3d_152(resnet3d_test):
        """Test 152."""
        K.set_image_data_format('channels_last')
>       model = Resnet3DBuilder.build_resnet_152((224, 224, 224, 1), 2)

tests/test_resnet3d.py:60: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
resnet3d/resnet3d.py:297: in build_resnet_152
    [3, 8, 36, 3], reg_factor=reg_factor)
resnet3d/resnet3d.py:243: in build
    )(block)
resnet3d/resnet3d.py:113: in f
    )(input)
resnet3d/resnet3d.py:169: in f
    return _shortcut3d(input, residual)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

input = <tf.Tensor 'max_pooling3d_4/MaxPool3D:0' shape=(None, 56, 56, 56, 64) dtype=float32>
residual = <tf.Tensor 'conv3d_17/BiasAdd:0' shape=(None, 56, 56, 56, 256) dtype=float32>

    def _shortcut3d(input, residual):
        """3D shortcut to match input and residual and merges them with "sum"."""
>       stride_dim1 = ceil(input._keras_shape[DIM1_AXIS] \
            / residual._keras_shape[DIM1_AXIS])
E       AttributeError: 'Tensor' object has no attribute '_keras_shape'

resnet3d/resnet3d.py:80: AttributeError
_______________________________________________________________________________ test_get_block ________________________________________________________________________________

    def test_get_block():
        """Test get residual block."""
        K.set_image_data_format('channels_last')
        Resnet3DBuilder.build((224, 224, 224, 1), 2, 'bottleneck',
>                             [2, 2, 2, 2], reg_factor=1e-4)

tests/test_resnet3d.py:78: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
resnet3d/resnet3d.py:243: in build
    )(block)
resnet3d/resnet3d.py:113: in f
    )(input)
resnet3d/resnet3d.py:169: in f
    return _shortcut3d(input, residual)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

input = <tf.Tensor 'max_pooling3d_5/MaxPool3D:0' shape=(None, 56, 56, 56, 64) dtype=float32>
residual = <tf.Tensor 'conv3d_21/BiasAdd:0' shape=(None, 56, 56, 56, 256) dtype=float32>

    def _shortcut3d(input, residual):
        """3D shortcut to match input and residual and merges them with "sum"."""
>       stride_dim1 = ceil(input._keras_shape[DIM1_AXIS] \
            / residual._keras_shape[DIM1_AXIS])
E       AttributeError: 'Tensor' object has no attribute '_keras_shape'

resnet3d/resnet3d.py:80: AttributeError
============================================================================== warnings summary ===============================================================================
../../../../home/thor/miniconda3/envs/keras3d/lib/python3.6/site-packages/tensorflow/python/autograph/utils/testing.py:21
  /home/thor/miniconda3/envs/keras3d/lib/python3.6/site-packages/tensorflow/python/autograph/utils/testing.py:21: DeprecationWarning: the imp module is deprecated in favour of importlib; see the module's documentation for alternative uses
    import imp

-- Docs: https://docs.pytest.org/en/stable/warnings.html

---------- coverage: platform linux, python 3.6.11-final-0 -----------
Name                   Stmts   Miss  Cover   Missing
----------------------------------------------------
resnet3d/__init__.py       1      0   100%
resnet3d/resnet3d.py     124     25    80%   82-99, 108, 114, 131, 157, 185-188, 195, 244-267
----------------------------------------------------
TOTAL                    125     25    80%

=========================================================================== short test summary info ===========================================================================
FAILED tests/test_resnet3d.py::test_resnet3d_18 - AttributeError: 'Tensor' object has no attribute '_keras_shape'
FAILED tests/test_resnet3d.py::test_resnet3d_34 - AttributeError: 'Tensor' object has no attribute '_keras_shape'
FAILED tests/test_resnet3d.py::test_resnet3d_50 - AttributeError: 'Tensor' object has no attribute '_keras_shape'
FAILED tests/test_resnet3d.py::test_resnet3d_101 - AttributeError: 'Tensor' object has no attribute '_keras_shape'
FAILED tests/test_resnet3d.py::test_resnet3d_152 - AttributeError: 'Tensor' object has no attribute '_keras_shape'
FAILED tests/test_resnet3d.py::test_get_block - AttributeError: 'Tensor' object has no attribute '_keras_shape'`

how to run this codes?

Hi,JihongJu
Is there a simple example about how to use this project including run and test resnet3D,thanks!
Best regards
Gu Yu

NaN loss when training ResNet3D

Hello,

I am encountering a problem of NaN loss when training your architecture on a dataset of videos (is it a symptom of Exploding/Vanishing gradients ?).
I used Adam(lr=1e-3) as an optimisation function.

To tackle this, I'm trying different approaches :

• Setting the kernel_initializer to default initialization (glorot_uniform distribution).
• Removing the l2 regularization and the reg_factor (even if I incur an overfitting problem).
  My next move if it doesn't tackle the problem :
• Changing the categorical cross entropy loss by a mean squared error.

What do you think about my problem ? Do you have any other suggestions ?

A multi-branch 3D ResNet based network, an ambigous global loss

Hi @JihongJu
Thank you for this repository,
I have tried to conceive a multi-branch network consisting of two branches, the first is a 3D ResNet50 and the second is 2D ResNet50. However, the global loss which is theorically a weighted summation of two partial losses, is much higher than the weighted summation. I have tested with other networks and it works correctly. Unfortunately, it is not the case with this code. Here is my code
https://github.com/Emnaa12/Multi-branch-CNN
and here some training epoch results:

Epoch 2/200
26/26 [==============================] - 39s 1s/step - loss: 9.2902 - 
dense_1_loss: 0.0801 - dense_2_loss: 0.0717 -
Epoch 3/200
26/26 [==============================] - 39s 1s/step - loss: 8.2261 - 
dense_1_loss: 0.0251 - dense_2_loss: 0.0199 -
Epoch 4/200 
26/26 [==============================] - 39s 2s/step - loss: 7.3107 - 
dense_1_loss: 0.0595 - dense_2_loss: 0.0048 

Could you please have a look to my code and figure out the key problem of this issue?

input_shape

Hey,

I realized that only certain input_shapes seem to be working or I get a broadcast error.
Operands could not be broadcast together with shapes

Do you know how I can easily fix this?
I am working with (19,19,19) cubes atm
These are the input_shapes that seem to be allowed:

(1,1,1) : ok
(2,2,2) : ok
(3,3,3) : ok
(4,4,4) : ok
(5,5,5) : ok
(6,6,6) : ok
(7,7,7) : ok
(8,8,8) : ok
(9,9,9) : Operands could not be broadcast together with shapes (3, 3, 3, 128) (2, 2, 2, 128)
(10,10,10) : Operands could not be broadcast together with shapes (3, 3, 3, 128) (2, 2, 2, 128)
(11,11,11) : Operands could not be broadcast together with shapes (3, 3, 3, 128) (2, 2, 2, 128)
12 : Operands could not be broadcast together with shapes (3, 3, 3, 512) (2, 2, 2, 512)
13 : ok
14 : ok
15 : ok
16 : ok
17 : Operands could not be broadcast together with shapes (5, 5, 5, 512) (3, 3, 3, 512)
18 : Operands could not be broadcast together with shapes (5, 5, 5, 512) (3, 3, 3, 512)
19 : Operands could not be broadcast together with shapes (5, 5, 5, 512) (3, 3, 3, 512)
20 : Operands could not be broadcast together with shapes (5, 5, 5, 512) (3, 3, 3, 512)
21 : Operands could not be broadcast together with shapes (3, 3, 3, 1024) (2, 2, 2, 1024)
22 : Operands could not be broadcast together with shapes (3, 3, 3, 1024) (2, 2, 2, 1024)
23 : Operands could not be broadcast together with shapes (3, 3, 3, 1024) (2, 2, 2, 1024)
24 : Operands could not be broadcast together with shapes (3, 3, 3, 1024) (2, 2, 2, 1024)
25 : Operands could not be broadcast together with shapes (7, 7, 7, 512) (4, 4, 4, 512)
26 : Operands could not be broadcast together with shapes (7, 7, 7, 512) (4, 4, 4, 512)
27 : Operands could not be broadcast together with shapes (7, 7, 7, 512) (4, 4, 4, 512)
28 : Operands could not be broadcast together with shapes (7, 7, 7, 512) (4, 4, 4, 512)
29 : ok
30 : ok
31 : ok
32 : ok
33 : Operands could not be broadcast together with shapes (9, 9, 9, 512) (5, 5, 5, 512)
34 : Operands could not be broadcast together with shapes (9, 9, 9, 512) (5, 5, 5, 512)
35 : Operands could not be broadcast together with shapes (9, 9, 9, 512) (5, 5, 5, 512)
36 : Operands could not be broadcast together with shapes (9, 9, 9, 512) (5, 5, 5, 512)
37 : Operands could not be broadcast together with shapes (5, 5, 5, 1024) (3, 3, 3, 1024)
38 : Operands could not be broadcast together with shapes (5, 5, 5, 1024) (3, 3, 3, 1024)
39 : Operands could not be broadcast together with shapes (5, 5, 5, 1024) (3, 3, 3, 1024)
40 : Operands could not be broadcast together with shapes (5, 5, 5, 1024) (3, 3, 3, 1024)
41 : Operands could not be broadcast together with shapes (11, 11, 11, 512) (6, 6, 6, 512)
42 : Operands could not be broadcast together with shapes (11, 11, 11, 512) (6, 6, 6, 512)
43 : Operands could not be broadcast together with shapes (11, 11, 11, 512) (6, 6, 6, 512)
44 : Operands could not be broadcast together with shapes (11, 11, 11, 512) (6, 6, 6, 512)
45 : Operands could not be broadcast together with shapes (3, 3, 3, 2048) (2, 2, 2, 2048)
46 : Operands could not be broadcast together with shapes (3, 3, 3, 2048) (2, 2, 2, 2048)
47 : Operands could not be broadcast together with shapes (3, 3, 3, 2048) (2, 2, 2, 2048)
48 : Operands could not be broadcast together with shapes (3, 3, 3, 2048) (2, 2, 2, 2048)
49 : Operands could not be broadcast together with shapes (13, 13, 13, 512) (7, 7, 7, 512)
50 : Operands could not be broadcast together with shapes (13, 13, 13, 512) (7, 7, 7, 512)
51 : Operands could not be broadcast together with shapes (13, 13, 13, 512) (7, 7, 7, 512)
52 : Operands could not be broadcast together with shapes (13, 13, 13, 512) (7, 7, 7, 512)
53 : Operands could not be broadcast together with shapes (7, 7, 7, 1024) (4, 4, 4, 1024)
54 : Operands could not be broadcast together with shapes (7, 7, 7, 1024) (4, 4, 4, 1024)
55 : Operands could not be broadcast together with shapes (7, 7, 7, 1024) (4, 4, 4, 1024)
56 : Operands could not be broadcast together with shapes (7, 7, 7, 1024) (4, 4, 4, 1024)
57 : Operands could not be broadcast together with shapes (15, 15, 15, 512) (8, 8, 8, 512)
58 : Operands could not be broadcast together with shapes (15, 15, 15, 512) (8, 8, 8, 512)
59 : Operands could not be broadcast together with shapes (15, 15, 15, 512) (8, 8, 8, 512)
60 : Operands could not be broadcast together with shapes (15, 15, 15, 512) (8, 8, 8, 512)
61 : ok
62 : ok
63 : ok
64 : ok
65 : Operands could not be broadcast together with shapes (17, 17, 17, 512) (9, 9, 9, 512)
66 : Operands could not be broadcast together with shapes (17, 17, 17, 512) (9, 9, 9, 512)
67 : Operands could not be broadcast together with shapes (17, 17, 17, 512) (9, 9, 9, 512)
68 : Operands could not be broadcast together with shapes (17, 17, 17, 512) (9, 9, 9, 512)
69 : Operands could not be broadcast together with shapes (9, 9, 9, 1024) (5, 5, 5, 1024)
70 : Operands could not be broadcast together with shapes (9, 9, 9, 1024) (5, 5, 5, 1024)
71 : Operands could not be broadcast together with shapes (9, 9, 9, 1024) (5, 5, 5, 1024)
72 : Operands could not be broadcast together with shapes (9, 9, 9, 1024) (5, 5, 5, 1024)
73 : Operands could not be broadcast together with shapes (19, 19, 19, 512) (10, 10, 10, 512)
74 : Operands could not be broadcast together with shapes (19, 19, 19, 512) (10, 10, 10, 512)
75 : Operands could not be broadcast together with shapes (19, 19, 19, 512) (10, 10, 10, 512)
76 : Operands could not be broadcast together with shapes (19, 19, 19, 512) (10, 10, 10, 512)
77 : Operands could not be broadcast together with shapes (5, 5, 5, 2048) (3, 3, 3, 2048)
78 : Operands could not be broadcast together with shapes (5, 5, 5, 2048) (3, 3, 3, 2048)
79 : Operands could not be broadcast together with shapes (5, 5, 5, 2048) (3, 3, 3, 2048)
80 : Operands could not be broadcast together with shapes (5, 5, 5, 2048) (3, 3, 3, 2048)
81 : Operands could not be broadcast together with shapes (21, 21, 21, 512) (11, 11, 11, 512)
82 : Operands could not be broadcast together with shapes (21, 21, 21, 512) (11, 11, 11, 512)
83 : Operands could not be broadcast together with shapes (21, 21, 21, 512) (11, 11, 11, 512)
84 : Operands could not be broadcast together with shapes (21, 21, 21, 512) (11, 11, 11, 512)
85 : Operands could not be broadcast together with shapes (11, 11, 11, 1024) (6, 6, 6, 1024)
86 : Operands could not be broadcast together with shapes (11, 11, 11, 1024) (6, 6, 6, 1024)
87 : Operands could not be broadcast together with shapes (11, 11, 11, 1024) (6, 6, 6, 1024)
88 : Operands could not be broadcast together with shapes (11, 11, 11, 1024) (6, 6, 6, 1024)
89 : Operands could not be broadcast together with shapes (23, 23, 23, 512) (12, 12, 12, 512)
90 : Operands could not be broadcast together with shapes (23, 23, 23, 512) (12, 12, 12, 512)
91 : Operands could not be broadcast together with shapes (23, 23, 23, 512) (12, 12, 12, 512)
92 : Operands could not be broadcast together with shapes (23, 23, 23, 512) (12, 12, 12, 512)
93 : ok
94 : ok
95 : ok
96 : ok
97 : Operands could not be broadcast together with shapes (25, 25, 25, 512) (13, 13, 13, 512)
98 : Operands could not be broadcast together with shapes (25, 25, 25, 512) (13, 13, 13, 512)
99 : Operands could not be broadcast together with shapes (25, 25, 25, 512) (13, 13, 13, 512)