ethanwharris/stawm

Code for the paper 'A Biologically Inspired Visual Working Memory for Deep Networks'

LinkedIn • Twitter • Google Scholar

FMix

• Code for our paper "FMix: Enhancing Mixed Sample Data Augmentation"
• Used by the second place team in the Bengali.AI Handwritten Grapheme Classification Kaggle competition and by the third place team in the Rainforest Connection Species Audio Detection Kaggle competition

---

Torchbearer

• A model training library for PyTorch
• Formerly part of the PyTorch ecosystem
• Used in the University of Southampton Differentiable Programming (and Deep Learning) module
• Joined PyTorch-Lightning last year

stawm's Issues

RuntimeError:

hello sir! I tried to run your code,but there are some bug and do not know how to solve,here are the Traceback :

MNIST:
File "/home/sdc/Desktop/STAWM-master/mnist_class_28.py", line 121, in
run(8, 512)
File "/home/sdc/Desktop/STAWM-master/mnist_class_28.py", line 115, in run
trial.run(200)
File "/home/sdc/Desktop/test/test/lib/python3.5/site-packages/torchbearer/trial.py", line 121, in wrapper
res = func(self, *args, **kwargs)
File "/home/sdc/Desktop/test/test/lib/python3.5/site-packages/torchbearer/trial.py", line 703, in run
final_metrics = self._fit_pass(state)[torchbearer.METRICS]
File "/home/sdc/Desktop/test/test/lib/python3.5/site-packages/torchbearer/trial.py", line 276, in wrapper
res = func(self, *args, **kwargs)
File "/home/sdc/Desktop/test/test/lib/python3.5/site-packages/torchbearer/trial.py", line 768, in _fit_pass
state[torchbearer.OPTIMIZER].step(closure)
File "/home/sdc/Desktop/test/test/lib/python3.5/site-packages/torch/optim/adam.py", line 58, in step
loss = closure()
File "/home/sdc/Desktop/test/test/lib/python3.5/site-packages/torchbearer/trial.py", line 752, in closure
state[torchbearer.Y_PRED] = state[torchbearer.MODEL](state[torchbearer.X], state=state)
File "/home/sdc/Desktop/test/test/lib/python3.5/site-packages/torch/nn/modules/module.py", line 493, in call
result = self.forward(*input, **kwargs)
File "/home/sdc/Desktop/STAWM-master/mnist_class_28.py", line 83, in forward
x = self.memory.glimpse(x, image)
File "/home/sdc/Desktop/STAWM-master/memory.py", line 112, in glimpse
x = self.locator(pose, image)
File "/home/sdc/Desktop/test/test/lib/python3.5/site-packages/torch/nn/modules/module.py", line 493, in call
result = self.forward(*input, **kwargs)
File "/home/sdc/Desktop/STAWM-master/modules.py", line 29, in forward
theta = theta.view(x.size(0), 2, 4)
RuntimeError: shape '[128, 2, 4]' is invalid for input of size 768

Celeba:
File "/home/sdc/Desktop/STAWM-master/celeba_draw_32.py", line 292, in
run(8, 32, 128, 0, device='cuda')
File "/home/sdc/Desktop/STAWM-master/celeba_draw_32.py", line 288, in run
trial.run(100)
File "/home/sdc/Desktop/test/test/lib/python3.5/site-packages/torchbearer/trial.py", line 121, in wrapper
res = func(self, *args, **kwargs)
File "/home/sdc/Desktop/test/test/lib/python3.5/site-packages/torchbearer/trial.py", line 703, in run
final_metrics = self._fit_pass(state)[torchbearer.METRICS]
File "/home/sdc/Desktop/test/test/lib/python3.5/site-packages/torchbearer/trial.py", line 276, in wrapper
res = func(self, *args, **kwargs)
File "/home/sdc/Desktop/test/test/lib/python3.5/site-packages/torchbearer/trial.py", line 768, in _fit_pass
state[torchbearer.OPTIMIZER].step(closure)
File "/home/sdc/Desktop/test/test/lib/python3.5/site-packages/torch/optim/adam.py", line 58, in step
loss = closure()
File "/home/sdc/Desktop/test/test/lib/python3.5/site-packages/torchbearer/trial.py", line 752, in closure
state[torchbearer.Y_PRED] = state[torchbearer.MODEL](state[torchbearer.X], state=state)
File "/home/sdc/Desktop/test/test/lib/python3.5/site-packages/torch/nn/modules/module.py", line 493, in call
result = self.forward(*input, **kwargs)
File "/home/sdc/Desktop/STAWM-master/celeba_draw_32.py", line 146, in forward
x, inverse = self.memory.glimpse(x, image)
File "/home/sdc/Desktop/STAWM-master/memory.py", line 112, in glimpse
x = self.locator(pose, image)
File "/home/sdc/Desktop/test/test/lib/python3.5/site-packages/torch/nn/modules/module.py", line 493, in call
result = self.forward(*input, **kwargs)
File "/home/sdc/Desktop/STAWM-master/modules.py", line 29, in forward
theta = theta.view(x.size(0), 2, 4)
RuntimeError: shape '[128, 2, 4]' is invalid for input of size 192

Request source code

I am very interested in your recent ICLR2019 work. When will you publish the source code?

The mnist.gif image

Hi,
I am wondering how do you plot the gif images. Mainly, how do you transfer the affine matrix to the red rhomboids in mnist.gif? It would be great if you kindly share the code.
Thank you very much!

Cheers,
Guozhang