dahiyaaneesh / peclr

This is the pretraining code for PeCLR. An equivariant contrastive learning framework for 3D hand pose estimation. The paper is presented at ICCV 2021.
https://ait.ethz.ch/projects/2021/PeCLR/
MIT License
83 stars 14 forks source link

visualize some images #15

Open chenzf11 opened 1 year ago

chenzf11 commented 1 year ago

Hello!your work is so interesting! I want to visualize some images, so I use your trained model "rn152_peclr_yt3d-fh_pt_fh_ft.pth" ,the function plot_hand in src/visualization/visualize.py, but I get some terrible results(neglect the axis problem just see the coord key point). How can I get the correct result like the paper show?Thank you for your reply! The following are my crazy result: 514244d01208872db23adc609f24f8f 7ea751f75c782cc7d5aac39e6d9de2c

eng100200 commented 1 year ago

@sysu19351018 Hello, I am having the same problem. Have you resolved it yet? if not lets discuss together.

shenpeiqi commented 1 year ago

@sysu19351018您好,我遇到了同样的问题。你解决了吗?如果没有,让我们一起讨论。

Hello, I also encountered the same problem, can you please solve it, can you discuss it

eng100200 commented 1 year ago

@shenpeiqi If you are taking about plot and visualizing the Key Points, Yes, I have resolved it. Do you have other than this Problem? result2023today22

shenpeiqi commented 1 year ago

@shenpeiqi If you are taking about plot and visualizing the Key Points, Yes, I have resolved it. Do you have other than this Problem? result2023today22

Thank you very much for your reply, I have implemented the part about hand key point visualization, which is to project three-dimensional points into two-dimensional coordinates and then visualize them. I don't know if you reproduced this project, do you remember that in the pred_fh.py code, you first get the initialized bounding box, the code just takes the center area of the picture, is this reasonable, if the hand is not located in the center of the picture at the beginning, then the detection fails, then the author gave this part of the code is only for the freihand this data set, then its generalization ability is not very weak, do I need to add a hand detector? Because I just started researching this, there are many questions I don't understand, and I look forward to your reply, thank you again!

eng100200 commented 1 year ago

@shenpeiqi what is your project? Yes, hand detector would be needed, while, Scale, and K matrix would be needed, if not generalized. Add my WeChat 13121340434.