705062791 / pgbig Goto Github PK
View Code? Open in Web Editor NEWCVPR 2022
CVPR 2022
作者你好,我尝试复现你的代码,GPU用的3090,但我发现在h3.6m数据集上训练50个epoch需要20多个小时,请问你的情况也是如此吗?我在想是不是我哪里设置错了或者是服务器出现了问题。
Hi
I notice that you claimed in your abstract that this method outperforms previous approaches by large margins. Although Table.1 has shown this superiority, one of the important issues I have to point out is that the real SOTA is Motion Prediction using Trajectory Cues. However, compared with this paper, your method has a noticeable performance drop on both Human3.6M and CMU-MoCap datasets. In this case, it is hard to make readers agree with the performance superiority of your method claimed in the paper.
Hi,
thanks for sharing the great work!
I have notice that you have used split 2 for both validation and testing in your code of main_h36m_3d.py and main_3dpw_3d.py
I doubt if it's unfair to use the same data for testing and validation
Thanks
Hi @705062791 ,
Thanks for your excellent work!
About the visualization, do you know how to generate the results as in Figure 5 in your paper?
Hi,
Thanks for sharing your work!
I have some problems with the structure of the T-DGCN as here.
As illustrated in Eq.4 in the paper, the intermediate result Y is left-multiplied by an adjacency matrix
While as in the code, it seems you multiply Y with a weight matrix weight_seq
.
I wonder if I am right and could you please explain something about it?
Thanks,
I notice the number of joints in paper is 22(32-10), but the figure 4 has 24 joints. And in the provided code, there are 32 points.
Hello, do you know what the frame rates for the CMU-mocap and 3dpw datasets after preprocessing are? It seems that the frame rate for 3dpw is not 25 fps, and the frame rate for cmu-mocap is downsampled by 2 but is the original frame rate 50 fps?
Thank you very much.
Hello,
your work is very interresting!
Is it possible to run the model only providing an input sequence of 10 poses and getting the 25 predicted output sequences from the model?
I only managed to run the model using a full trainings/evaluation dataset but not using a single input sequence.
Thank you for your help and insights.
I notice that in line 273~276 of the file "main_cmu_3d.py", p3d_out_4 is treated as follow:
p3d_out_4 = p3d_h36.clone()[:, in_n:in_n + out_n]
p3d_out_4[:, :, dim_used] = p3d_out_all_4[:, seq_in:]
p3d_out_4[:, :, index_to_ignore] = p3d_out_4[:, :, index_to_equal]
p3d_out_4 = p3d_out_4.reshape([-1, out_n, all_dim//3, 3])
Will this result in some joints being counted multiple times when calculating MPJPE?
Thanks
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.