jhjacobsen / fully-invertible-revnet

fully-invertible-revnet

Not well documented and provided as is. This code is based on https://github.com/pclucas14/pytorch-glow

fully-invertible-revnet's Issues

Which paper is this repo for?

What paper is this repo for? Is it for Residual Flows for Invertible Generative Modeling?

train.py does not work on my machine

Hi,

I ran train.py (as is, without any modifications) with Python 3.7.7, Torch 1.1.0 and Ubuntu 18.04.2 LTS. It gave the following error:

Traceback (most recent call last): File "train.py", line 176, in <module> logits_ac = ac(_zn.detach()) File "/home/harshay/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 493, in __call__ result = self.forward(*input, **kwargs) File "/home/harshay/anaconda3/lib/python3.7/site-packages/torch/nn/modules/container.py", line 92, in forward input = module(input) File "/home/harshay/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 493, in __call__ result = self.forward(*input, **kwargs) File "/home/harshay/anaconda3/lib/python3.7/site-packages/torch/nn/modules/linear.py", line 92, in forward return F.linear(input, self.weight, self.bias) File "/home/harshay/anaconda3/lib/python3.7/site-packages/torch/nn/functional.py", line 1406, in linear ret = torch.addmm(bias, input, weight.t()) RuntimeError: size mismatch, m1: [128 x 3840], m2: [3072 x 512] at /opt/conda/conda-bld/pytorch_1556653215914/work/aten/src/THC/generic/THCTensorMathBlas.cu:268

Is there a way to fix this?

Thanks

Interpolation of nuisance variable

Hi,

Thanks for releasing this amazing library! I am trying to reproduce the Figure 7 in the paper. But when I interpolate the nuisance and compute the reverse, it seems model trained with cross entropy also behaves pretty well (the reverse sample is not dominated by nuisance). I notice that the logit is z[:,:10,2,2], do I just permute the rest of the elements to get figure 7?

Want to doublecheck if my implementation is correct.

A query : when can be expect the codes to be posted here?

when can we expect the codes to be posted here?

Just wanted check the implementation of DCT and what 'directly readout the logits from the DCT lowpass coefficients'.

Regards,
@sobalgi.