Comments (6)
Hi Saurav,
Thanks for contributing!
Pretty much this whole library is designed around TF1 constructs and in order to get it working with Keras and proper TF2 it probably needs a well thought out wholistic approach. We're also not ready to drop TF1 support, so we either need to have 2 versions of the code or else a way to keep things backwards compatible. The GANModel data structure, which is sort of the center of the whole library relies heavily on variable_scopes, which don't exist in TF2, so that might be the first thing to figure out, and then figure out how to adapt everything else from there.
Would you be interested in coming up with a plan of that sort, and then maybe a proof-of-concept?
The particular issue you point out here could be just one step of the larger plan.
from gan.
Hey @aaronsarna, that's a great suggestion. I was looking into the variable_scopes
as well while attempting to redefine the wasserstein_loss
function. I'll get started on building up a plan for converting v1.get_variable
to tf.Variable
and similarly to convert v1.variable_scope
to other keras objects like layers/models/module
from gan.
@aaronsarna do you want me to use a template for the Proof-of-concept?
from gan.
I think just showing a basic mnist or cifar10 GAN trained with TF2 and Keras is sufficient.
from gan.
The GANModel class uses the collections module for creating tuple subclasses with named fields. But as per the docs page, this module has been "Deprecated since version 3.3, will be removed in version 3.10". Would you advise me to use the collections.abc module instead or did you have another module in mind?
from gan.
I don't think there's any need to stick with collections.namedtuple. To some extent the right answer will depend on the overall design. It may just be a class, or you could look at using attr or dataclasses. Whatever makes the most sense given the overall design.
from gan.
Related Issues (20)
- Why this module is written in Estimator and not in keras? HOT 2
- Memory leak in eval functions
- Spinoff Tensorflow GAN does not work in 2.X version HOT 1
- Computation of Sliced Wasserstein Distance numerically unstable HOT 1
- On all examples I get the error: " doesn't work when executing eagerly" HOT 1
- Getting error "`tfgan.gan_model` doesn't work when executing eagerly." on cifar example HOT 1
- Tutorial not working in Colab HOT 4
- Not working HOT 4
- StartGAN drive download issue HOT 2
- How to apply identity loss in cyclegan model?
- L1 loss problem with GAN if some data has ground-truth and some not HOT 2
- How to use the Virtual Batch Normalization class? HOT 3
- Update to use keras? HOT 2
- Could you create a tutorial for training cyclegan o TPU? HOT 1
- Pix2Pix?
- Issue with tfgan_on_tpus.ipynb in the train step
- Performance issue in the definition of _sliced_wasserstein, tensorflow_gan/python/eval/sliced_wasserstein.py
- GAN models for generating HD images HOT 1
- Incompatibility with tensorflow 2.16 HOT 2
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from gan.