frozenburning / relighting4d Goto Github PK
View Code? Open in Web Editor NEW[ECCV 2022] Relighting4D: Neural Relightable Human from Videos
Home Page: https://frozenburning.github.io/projects/relighting4d/
License: Other
[ECCV 2022] Relighting4D: Neural Relightable Human from Videos
Home Page: https://frozenburning.github.io/projects/relighting4d/
License: Other
I directly download your code and run them. However, the process seems to stay at this line
Hello ! Thank you for your amazing work.
I see you are adding torch.ones to the light on the following light if "fix_light" is set to True.
Why is it there and isn't this going to make the new light brighter than it should ?
Hello again :)
I am looking into the BRDF you are using.
Why is alpha = roughness squared (gamma(x) ^2) in your work ?
The original paper https://www.graphics.cornell.edu/~bjw/microfacetbsdf.pdf
just takes alpha values without squaring them.
Is it just a way to remove negative roughness when training ?
Hi Zhaoxi,
Thanks for your great work!
It seems like zju-mocap and people-snapshot use different light settings. I'm confused about that, could you please explain that?
if 'xyzc' in cfg.exp_name or 'zju' in cfg.exp_name:
print("zju-mocap:{}, using different lighting......".format(cfg.exp_name))
lxyz[...,1] = -lxyz[...,1]
lxyz[...,0] = -lxyz[...,0]
else:
lxyz = lxyz[..., [0, 2, 1]]
lxyz[...,1] = -lxyz[...,1]
If I wanna train the model on my own dataset, which one should I choose?
Thanks in advance!
Hi, i downloaded some hdr from poly heaven. But i find some differences between the hdr downloaded and the hdr provided. The max value of the sunset i download is about 30000, but the provide hdr is about 80. Is there any pre-process while dealing with the new hdr map?
Thank you for your amazing work again! @FrozenBurning
I have another question regarding relighting the model from a novel camera view. I noticed that I can only relight the model from the camera used to train. If I want to relight the model from a novel camera view, what should I do? Could you please provide me with some advice? Thank you very much!
We use blender and smpl-x self-built dataset and how to get, Normal, Occlusion, Diffuse, Specular informations.
Hi Zhaoxi,
Thanks for the great work.
It seem that the code and configuration files to handle multi-view data like ZJU Mocap are missing now, will they be released as well?
Thank you very much for your great work! @FrozenBurning
I am trying to train the model using the custom dataset provided in the SynthesisHuman Dataset. As shown in the README.md file in the custom folder, I can ' export the SMPL parameters and SMPL vertices to two directories: params
and vertices
'. I already have the 'model.npz' file containing the SMPL parameters, which contains the following keys: minimal_shape, betas, Jtr_posed, bone_transforms, trans, root_orient, pose_body, pose_hand. However, I do not know how to export the SMPL parameters or vertices from this file. Could you please give me some hints to export the corresponding parameters?
Thank you very much!
Hello, happy for your good job. I have a question if only training in single view video, the camera pose ωo is fixed. Therefore, I can only relight human in one camera pose ωo.
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.