-
Notifications
You must be signed in to change notification settings - Fork 2
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
A couple of fixes and generative applications #3
Comments
Hi @jbmaxwell , Yes your fixes are right---I messed up a few things when merging the code of different experiments. Since MultiresConv is just a sequence modeling layer, it is applicable to any methods/tasks that need a sequence-to-sequence mapping. You could certainly do conditional generation in this AR model example (which is no different from other AR models) and "infilling"/diffusion type methods (where the sequence to sequence map is used for one step of infilling. Thanks for your interest in our work! |
Thanks for the speedy reply! By "there's a way", I'm guessing you mean to generate images from the output? I see that the input shape is (64, 3, 1024), which I assume is a batch of images reshaped to three "sequences" of pixels (R, G, and B sequences, I mean). The output is (64, 100, 1024), which I see is the output of the decoder... So yes, if you could put together a quick example, or even point me in the right direction, that would be great! |
Hi @jbmaxwell , Please see autoregressive_eval.py for an example of generating unconditional samples from the trained model. |
I'm hitting an error:
I can see that (Just as an initial test, I first trained on 50 epochs of UPDATE: Actually, although the init takes that argument, I see it doesn't store it, so the module can't provide it in the |
Hi @jbmaxwell , Yep the solution is just storing that in the class! |
I updated the code. Please let me know if there are any other issues! |
Actually, I was curious about training on different shapes. I did some work on trying to get non-image data (shape = (bs, 32, 512)) working, but kept hitting errors in the loss calculation. |
First of all, congratulations on what looks like very exciting work!
I wanted to check your generative modelling application by running
autoregressive.py
, however, I had to change a couple of things to get it running:mixing
argument. I figured, since it was being givenFalse
as an input, I could probably just comment it out, which does run.forward()
function callsout = self.output_mapping(x)
, since that function doesn't seem to exist. I guessed this should probably beout = self.decoder(x)
, which seems to be running.... and converging. :)However, if I'm wrong about these "fixes" please let me know!
More generally, I'm very curious about the application of this model in generative contexts. I'm specifically in the music and audio field, where powerful sequence models are (obviously) essential.
So, a few things I'm wondering about:
autoregressive_eval.py
script, so I can see the output? (I can obviously dig in and figure this out, but if there's a quick mod you can suggest that would be great).Again, thanks for your work.
The text was updated successfully, but these errors were encountered: