Added an AlexNet wrapper using LRP layer definitions. #1
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…xpected with older code)
…port in convs); also add __init__.py files to allow for adding interprettensor to /data/ruthfong/tensorflow/:/data/ruthfong/tensorflow/models/slim/nets:/data/ruthfong/tensorflow/models:/data/ruthfong/tensorflow/cleverhans:/data/ruthfong/tensorflow/interprettensor::/users/ruthfong/sample_code/Caffe-ExcitationBP/python and then importing sub-folders as modules.
…sn't work (doberman.png), and helper file with imagenet classes
VigneshSrinivasan10
approved these changes
Jul 13, 2017
VigneshSrinivasan10
requested changes
Jul 13, 2017
VigneshSrinivasan10
requested changes
Jul 13, 2017
| self.strides = [1,self.stride_size, self.stride_size,1] | ||
| with tf.variable_scope(self.name): | ||
| self.weights = variables.weights(self.weights_shape) | ||
| self.biases = variables.biases(self.output_depth) | ||
| if self.weights is None: |
There was a problem hiding this comment.
Also, I am not exactly sure why you are doing this.
Though, the recent update for weight initialization should take care of this.
There was a problem hiding this comment.
Before, I wasn't able to set the weights to pre-trained weights after initialization, so I moved that logic to the point where we directly initialize the weights Variable. I'll take a look at the recent update though.
There was a problem hiding this comment.
I am not sure if having 'group' kernels helps in any way. The original AlexNet paper did it when they had problems with memory while computing.
BVLC/caffe#778
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
modules/convolution.py was modified to support a) multiple kernel groups, per AlexNet's original architecture, and b) initialization of weights/biases with passed-in arrays (older code should still work).
models/alexnet.py allows pre-trained weights to be loaded.
(Weights can be downloaded from http://www.cs.toronto.edu/~guerzhoy/tf_alexnet/bvlc_alexnet.npy)
Note: This implementation is based on https://github.com/guerzh/tf_weights, which doesn't work perfectly / seems to do well on some images but not others,
as noted at https://github.com/guerzh/tf_weights/issues/.
examples/alexnet_demo.py
I've included demo code and two example images that demonstrate a correctly behaving (examples/poodle.png) and incorrectly behaving (examples/doberman.png) classification result.
__init__.pyandmodels/__init__.pyThis allows
/path/to/interprettensorto be exported to PYTHONPATH and modules and models to be called as done in examples/alexnet_demo.pyExample:
export PYTHONPATH="/data/ruthfong/tensorflow/interprettensor":$PYTHONPATH