An error cast when I run train.py
JFChi opened this issue · comments
Dear Edward:
Thanks for your sharing. But when I run train.py, it cast our an error:ValueError: Filter must not be larger than the input: Filter: (3, 1) Input: (1, 80).
Do you know why?
Hi, could you post launch command and entire python error, please ?
I ran into this same problem, I think it has to do with the dimension ordering in images for tensorflow vs. theano. I fixed it by explicitly setting the backend to theano:
export KERAS_BACKEND=theano
I did run into more dimension problems later in the merge layers after fixing this though...
Year, kramimus, right, thanks a lot.
I configure my keras in the configuration file ~/.keras/keras.json
cat ~/.keras/keras.json
{
"image_dim_ordering": "th",
"epsilon": 1e-07,
"floatx": "float32",
"backend": "theano"
}
I've updated readme to address this issue.
Hi, I got the same type of error even though I setup the backend as tensorflow, and use the image_dim_ordering as tf.
It seems to me that the problem comes from here def get_unet(): inputs = Input((1,img_rows, img_cols))
I tried both this way and inputs = Input((img_rows, img_cols,1))
. Neighther solves the problem. Any insight would be appreciated.
Traceback (most recent call last):
File "train.py", line 155, in <module>
train_and_predict()
File "train.py", line 121, in train_and_predict
model = get_unet()
File "train.py", line 36, in get_unet
conv1 = Convolution2D(32, 3, 3, activation='relu', border_mode='same')(inputs)
File "/home/tfw/lib/python3.4/site-packages/Keras-1.0.3-py3.4.egg/keras/engine/topology.py", line 485, in __call__
self.add_inbound_node(inbound_layers, node_indices, tensor_indices)
.......
File "/home/tfw/lib/python3.4/site-packages/tensorflow/python/ops/common_shapes.py", line 238, in conv2d_shape
in_rows, in_cols, filter_rows, filter_cols, stride_r, stride_c, padding)
File "/home/tfw/lib/python3.4/site-packages/tensorflow/python/ops/common_shapes.py", line 151, in get2d_conv_output_size
% (filter_height, filter_width, input_height, input_width))
ValueError: filter must not be larger than the input: Filter: [3x3] Input: [64x1]
I can't comment issues regarding TF, code works with theano, although I managed to launch with TF backend, but got 10x slower training iteration.