kitti_train.py cannot initialize weights from downloaded data

[arvoelke]

Hi, I am currently trying to run your model using Python 2.7 and Theano==0.9.0 (using a GPU with pygpu==0.6.2). I am skipping the first step (by executing download_data.sh) and going straight to training (python kitti_train.py).

I initially tried using keras==2.0.2, but saw this error:

Traceback (most recent call last):
  File "kitti_train.py", line 59, in <module>
    errors = prednet(inputs)  # errors will be (batch_size, nt, nb_layers)
  File "/home/arvoelke/.virtualenvs/CTN/local/lib/python2.7/site-packages/keras/layers/recurrent.py", line 252, in __call__
    return super(Recurrent, self).__call__(inputs, **kwargs)
  File "/home/arvoelke/.virtualenvs/CTN/local/lib/python2.7/site-packages/keras/engine/topology.py", line 528, in __call__
    self.build(input_shapes[0])
  File "/home/arvoelke/prednet/prednet.py", line 216, in build
    if self.initial_weights is not None:
AttributeError: 'PredNet' object has no attribute 'initial_weights'

Then I noticed you suggest keras==1.2.0, so I downgraded, and retried, but saw a different error:

python kitti_train.py
Using Theano backend.
Can not use cuDNN on context None: Device not supported
Mapped name None to device cuda: Tesla C2075 (0000:02:00.0)
Traceback (most recent call last):
  File "kitti_train.py", line 60, in <module>
    errors_by_time = TimeDistributed(Dense(1, weights=[layer_loss_weights, np.zeros(1)], trainable=False), trainable=False)(errors)  # calculate weighted error by layer
  File "/home/arvoelke/.virtualenvs/CTN/local/lib/python2.7/site-packages/keras/engine/topology.py", line 543, in __call__
    self.build(input_shapes[0])
  File "/home/arvoelke/.virtualenvs/CTN/local/lib/python2.7/site-packages/keras/layers/wrappers.py", line 98, in build
    self.layer.build(child_input_shape)
  File "/home/arvoelke/.virtualenvs/CTN/local/lib/python2.7/site-packages/keras/layers/core.py", line 763, in build
    self.set_weights(self.initial_weights)
  File "/home/arvoelke/.virtualenvs/CTN/local/lib/python2.7/site-packages/keras/engine/topology.py", line 966, in set_weights
    str(weights)[:50] + '...')
ValueError: You called `set_weights(weights)` on layer "dense_1" with a  weight list of length 2, but the layer was expecting 0 weights. Provided weights: [array([[ 1.],
       [ 0.],
       [ 0.],
       ...

Any tips for things I could try? Thanks, Aaron

[arvoelke]

I downgraded even further to keras==1.0.8, and now it seems to be working. This version should probably be suggested in the README?

[bill-lotter]

Hi, sorry about the versioning trouble. I'm going to try to make it compatible with Keras 2.0 now, so all of this might be moot point, but it looks like I was using Keras 1.2.0 a few commits ahead of the official 1.2.0 release (specifically I was on 3a7cd05b488e327c1adccaaff10c78390d53b5a8) and it worked on that commit and 1.0.8 but not on the official 1.2.0 commit in-between. Sorry about that! Keras 2.0 is supposed to be stable release, so hopefully there won't be these issues in the future once I finish the upgrade.

[dendisuhubdy]

@arvoelke got the same error here, I'm trying your solution to downgrade Keras

(lisa) suhubdyd@grok-machine:~/research/prednet$ python kitti_train.py 
Using Theano backend.
Using cuDNN version 5105 on context None
Mapped name None to device cuda: GeForce GTX 750 (0000:01:00.0)
/u/suhubdyd/research/prednet/prednet.py:187: UserWarning: Update your `Conv2D` call to the Keras 2 API: `Conv2D(3, (3, 3), padding="same", activation=<function ..., data_format="channels_last")`
  self.conv_layers[c].append(Convolution2D(self.R_stack_sizes[l], self.R_filt_sizes[l], self.R_filt_sizes[l], border_mode='same', activation=act, dim_ordering=self.dim_ordering))
/u/suhubdyd/research/prednet/prednet.py:190: UserWarning: Update your `Conv2D` call to the Keras 2 API: `Conv2D(3, (3, 3), padding="same", activation="relu", data_format="channels_last")`
  self.conv_layers['ahat'].append(Convolution2D(self.stack_sizes[l], self.Ahat_filt_sizes[l], self.Ahat_filt_sizes[l], border_mode='same', activation=act, dim_ordering=self.dim_ordering))
/u/suhubdyd/research/prednet/prednet.py:193: UserWarning: Update your `Conv2D` call to the Keras 2 API: `Conv2D(48, (3, 3), padding="same", activation=<function ..., data_format="channels_last")`
  self.conv_layers['a'].append(Convolution2D(self.stack_sizes[l+1], self.A_filt_sizes[l], self.A_filt_sizes[l], border_mode='same', activation=self.A_activation, dim_ordering=self.dim_ordering))
/u/suhubdyd/research/prednet/prednet.py:187: UserWarning: Update your `Conv2D` call to the Keras 2 API: `Conv2D(48, (3, 3), padding="same", activation=<function ..., data_format="channels_last")`
  self.conv_layers[c].append(Convolution2D(self.R_stack_sizes[l], self.R_filt_sizes[l], self.R_filt_sizes[l], border_mode='same', activation=act, dim_ordering=self.dim_ordering))
/u/suhubdyd/research/prednet/prednet.py:190: UserWarning: Update your `Conv2D` call to the Keras 2 API: `Conv2D(48, (3, 3), padding="same", activation=<function ..., data_format="channels_last")`
  self.conv_layers['ahat'].append(Convolution2D(self.stack_sizes[l], self.Ahat_filt_sizes[l], self.Ahat_filt_sizes[l], border_mode='same', activation=act, dim_ordering=self.dim_ordering))
/u/suhubdyd/research/prednet/prednet.py:193: UserWarning: Update your `Conv2D` call to the Keras 2 API: `Conv2D(96, (3, 3), padding="same", activation=<function ..., data_format="channels_last")`
  self.conv_layers['a'].append(Convolution2D(self.stack_sizes[l+1], self.A_filt_sizes[l], self.A_filt_sizes[l], border_mode='same', activation=self.A_activation, dim_ordering=self.dim_ordering))
/u/suhubdyd/research/prednet/prednet.py:187: UserWarning: Update your `Conv2D` call to the Keras 2 API: `Conv2D(96, (3, 3), padding="same", activation=<function ..., data_format="channels_last")`
  self.conv_layers[c].append(Convolution2D(self.R_stack_sizes[l], self.R_filt_sizes[l], self.R_filt_sizes[l], border_mode='same', activation=act, dim_ordering=self.dim_ordering))
/u/suhubdyd/research/prednet/prednet.py:190: UserWarning: Update your `Conv2D` call to the Keras 2 API: `Conv2D(96, (3, 3), padding="same", activation=<function ..., data_format="channels_last")`
  self.conv_layers['ahat'].append(Convolution2D(self.stack_sizes[l], self.Ahat_filt_sizes[l], self.Ahat_filt_sizes[l], border_mode='same', activation=act, dim_ordering=self.dim_ordering))
/u/suhubdyd/research/prednet/prednet.py:193: UserWarning: Update your `Conv2D` call to the Keras 2 API: `Conv2D(192, (3, 3), padding="same", activation=<function ..., data_format="channels_last")`
  self.conv_layers['a'].append(Convolution2D(self.stack_sizes[l+1], self.A_filt_sizes[l], self.A_filt_sizes[l], border_mode='same', activation=self.A_activation, dim_ordering=self.dim_ordering))
/u/suhubdyd/research/prednet/prednet.py:187: UserWarning: Update your `Conv2D` call to the Keras 2 API: `Conv2D(192, (3, 3), padding="same", activation=<function ..., data_format="channels_last")`
  self.conv_layers[c].append(Convolution2D(self.R_stack_sizes[l], self.R_filt_sizes[l], self.R_filt_sizes[l], border_mode='same', activation=act, dim_ordering=self.dim_ordering))
/u/suhubdyd/research/prednet/prednet.py:190: UserWarning: Update your `Conv2D` call to the Keras 2 API: `Conv2D(192, (3, 3), padding="same", activation=<function ..., data_format="channels_last")`
  self.conv_layers['ahat'].append(Convolution2D(self.stack_sizes[l], self.Ahat_filt_sizes[l], self.Ahat_filt_sizes[l], border_mode='same', activation=act, dim_ordering=self.dim_ordering))
/u/suhubdyd/research/prednet/prednet.py:195: UserWarning: Update your `UpSampling2D` call to the Keras 2 API: `UpSampling2D(data_format="channels_last")`
  self.upsample = UpSampling2D(dim_ordering=self.dim_ordering)
/u/suhubdyd/research/prednet/prednet.py:196: UserWarning: Update your `MaxPooling2D` call to the Keras 2 API: `MaxPooling2D(data_format="channels_last")`
  self.pool = MaxPooling2D(dim_ordering=self.dim_ordering)
Traceback (most recent call last):
  File "kitti_train.py", line 59, in <module>
    errors = prednet(inputs)  # errors will be (batch_size, nt, nb_layers)
  File "/u/suhubdyd/.conda/envs/lisa/lib/python2.7/site-packages/keras/layers/recurrent.py", line 243, in __call__
    return super(Recurrent, self).__call__(inputs, **kwargs)
  File "/u/suhubdyd/.conda/envs/lisa/lib/python2.7/site-packages/keras/engine/topology.py", line 558, in __call__
    self.build(input_shapes[0])
  File "/u/suhubdyd/research/prednet/prednet.py", line 216, in build
    if self.initial_weights is not None:
AttributeError: 'PredNet' object has no attribute 'initial_weights'

[xukai8]

@bill-lotter @arvoelke I got the same problem， I use the keras==2.0.6 and the tensorflow backend.

Traceback (most recent call last): File "/home/xk/PycharmProjects/prednet/kitti_train.py", line 59, in <module> errors = prednet(inputs) # errors will be (batch_size, nt, nb_layers) File "/home/xk/anaconda3/envs/tf2/lib/python2.7/site-packages/keras/layers/recurrent.py", line 262, in __call__ return super(Recurrent, self).__call__(inputs, **kwargs) File "/home/xk/anaconda3/envs/tf2/lib/python2.7/site-packages/keras/engine/topology.py", line 569, in __call__ self.build(input_shapes[0]) File "/home/xk/PycharmProjects/prednet/prednet.py", line 216, in build if self.initial_weights is not None: AttributeError: 'PredNet' object has no attribute 'initial_weights'

[bill-lotter]

See cc76248 for downloading weights that are compatible with Keras 2.0. You can also convert old PredNet weights for Keras 2.0 by using convert_model_to_keras2 in keras_utils.py.