search

DmitryUlyanov / texture_nets

Code for "Texture Networks: Feed-forward Synthesis of Textures and Stylized Images" paper.
Apache License 2.0
1.22k stars 217 forks source link

All brushes same size #28

Open Heartsie opened 8 years ago

Heartsie commented 8 years ago

Hi! It looks like there is a problem where all the "brush strokes" in a style are all the same size. You can really see it here: karya-oil_07-stylweight3-stylesize512-learning0-001-model 40000 This was the style image I used: oil_07

40,000 iterations of the microsoft coco dataset. styleweight of 3, stylesize of 512, image size of 512, learning set to .001, batch of 1. I got the same results with different style weights and learning sizes. Unfortunately this looks like a big limitation :(

lancodec commented 8 years ago

@rogozin70 what other parameters? tv_weight and learning_rate?

rogozin70 commented 8 years ago

tv_weight 0.000085, learning_rate 0.001

SantaFlamel commented 8 years ago

normalization?

rogozin70 commented 8 years ago

-normalize_gradients', 'true'

rogozin70 commented 8 years ago

What parameter - bathch size? what happens if i put batch size - 2 or 3, 4?

rogozin70 commented 8 years ago

plastelin Plasticine person

ycjing commented 6 years ago

A recent work aims to exploit one single model to achieve continuous stroke size control in different output images or distinct stroke sizes in different spatial regions within the same output image. The project page: http://yongchengjing.com/StrokeControllable
Demo video: https://youtu.be/UNG38tdMSMg

3DTOPO commented 6 years ago

@ycjing do you have an implementation of your paper we can take a look at?

ycjing commented 6 years ago

@3DTOPO Thanks for your interest. We are organizing our code these days and will release our code as well as all the pre-trained models soon. I will post the link in this issue when the code is ready. I would also like to recommend our review for anyone who is interested in Neural Style Transfer: https://github.com/ycjing/Neural-Style-Transfer-Papers Thanks!

ycjing commented 6 years ago

Hi, @3DTOPO Our code and pre-trained models are finally ready: https://github.com/LouieYang/stroke-controllable-fast-style-transfer Also, we have updated our paper correspondingly at: https://arxiv.org/abs/1802.07101 Thanks.