own dataset: Shapes (64, 4, 4, 256) and (64, 3, 3, 256) are not compatible

Hi there,

When using my own dataset (28x28, grayscale), I get an error for this line in model.py:

g_optim = tf.train.AdamOptimizer(config.learning_rate, beta1=config.beta1) \ .minimize(self.g_loss, var_list=self.g_vars)

error: Traceback (most recent call last): File "...tensorflow/python/framework/tensor_shape.py", line 575, in merge_with new_dims.append(dim.merge_with(other[i])) File "...tensorflow/python/framework/tensor_shape.py", line 133, in merge_with self.assert_is_compatible_with(other) File "...tensorflow/python/framework/tensor_shape.py", line 108, in assert_is_compatible_with % (self, other)) ValueError: Dimensions 4 and 3 are not compatible During handling of the above exception, another exception occurred: Traceback (most recent call last): File "main.py", line 74, in <module> tf.app.run() File "...tensorflow/python/platform/app.py", line 30, in run sys.exit(main(sys.argv[:1] + flags_passthrough)) File "main.py", line 59, in main dcgan.train(FLAGS) File "...DCGAN-tensorflow/model.py", line 139, in train .minimize(self.g_loss, var_list=self.g_vars) File "...tensorflow/python/training/optimizer.py", line 196, in minimize grad_loss=grad_loss) File "...tensorflow/python/training/optimizer.py", line 253, in compute_gradients colocate_gradients_with_ops=colocate_gradients_with_ops) File "...tensorflow/python/ops/gradients.py", line 491, in gradients in_grad.set_shape(t_in.get_shape()) File "...tensorflow/python/framework/ops.py", line 408, in set_shape self._shape = self._shape.merge_with(shape) File "...tensorflow/python/framework/tensor_shape.py", line 579, in merge_with (self, other)) ValueError: Shapes (64, 4, 4, 256) and (64, 3, 3, 256) are not compatible

Any advice on how to fix this?

[carpedm20]

@hercky Oh I see. The problem was the size of input is too small for 4 layer networks and it’s not dividable. I think it’s tricky problem that a depth of network highly depends on input width, height and there is no one answer for it. I’ll just add a raise Exception code if the input size is “too small” for the current network design and suggest people to change it.

[carpedm20]

I’m not sure but the fast simple bypass might be adding tf.image.resize_images right after

self.images = tf.placeholder(tf.float32, 
                            [self.batch_size] + [self.output_size, self.output_size, self.c_dim],
                            name='real_images')

is defined.