Generative Adversarial Networks (GANs) has achieved good performance in image style transfer, e.g. transfer a landscape photograph to a painting style (Monet, Van Gogh), and vice versa [1].
Should we expect the same good performance for using GANs to tranfer a textureless 3D model image to a real-life photo style? If so, what is the general implementation procedure, and do I need to train the GAN model from scatch with a great amount of images from both domains (3D model and real-life photo)?
Reference
[1] Zhu, Jun-Yan, et al. "Unpaired image-to-image translation using cycle-consistent adversarial networks." Proceedings of the IEEE international conference on computer vision. 2017.
Sachin Kumar, could you please tell me how i can realize it? Do I need a lot of data from both domains for training the model?
Thank you.
Hi do you know what is a 3D model with quads, triangles and so on ? Do you know what are U,V coordinates ?
Ya it is absolutely possible try using Enscape its trial is free for 1 Month.
Ahmed Khan ，could you briefly walk me througn how the plugin works?
I have just watched serveral tutorial videos, and it seems to me that what the software does is to dynamically render a 3D model with some prebuilt material libraries (e.g. using 3D grass material to render grass in this tutorial https://enscape3d.com/videotutorials/ )
What I am looking into is: given a buntch of real-life photos of the 3D model, can I render the 3D model with the style and texture of these photos? (please refer to the attached images)
Thank you~
Yes some exemples : https://stackoverflow.com/questions/2488086/texture-mapping-in-matlab
Sachin Kumar , I think MATLAB is just a tool for realizing many different kinds of algorithms. If a certain algorithm works for my case, I wouldn't mind which kind of tool to use, e.g. MATLAB, pyhon, tensorflow...
- Badges
- Science topic
- Similar topics
- Computer Science
- Images