Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

About 2D pose #34

Open
MilkLoo opened this issue Mar 13, 2024 · 15 comments
Open

About 2D pose #34

MilkLoo opened this issue Mar 13, 2024 · 15 comments

Comments

@MilkLoo
Copy link

MilkLoo commented Mar 13, 2024

Hello, according to the paper, I can see that the image has 2Dpose input, but in the code, I don't see where the 2Dpose input comes from. Could you please help me understand the following code?

@hongsukchoi
Copy link
Owner

Hi!

In the demo code, you can find the source here:

pose2d_result_path = './input/2d_pose_result.json'

In the dataloader code,

you can look at the def load_data(self): function.

@MilkLoo
Copy link
Author

MilkLoo commented Mar 15, 2024

7a6875956a75d2f536fbbae3224ba5f
Hello, I want to ask, I use your trained weight files and pictures, demo.py, but the effect is not so good, is what I did wrong?

@hongsukchoi
Copy link
Owner

Could you visualize your 2D pose input and full error log?

@MilkLoo
Copy link
Author

MilkLoo commented Mar 15, 2024

49636e1cb7cdd9322d1b04e5ed45f92
2dd5b60b8a4f0e4bc761b1339519583
I follow the operation of Redame.md step by step, and input is also downloaded from the link.

@hongsukchoi
Copy link
Owner

I see. Which link did you use?

And I think you can solve this problem by tuning the parameter here:

pose_thr = 0.1

The input 2D pose has an inaccurate lower body pose. You can filter those joints by increasing the threshold.

@MilkLoo
Copy link
Author

MilkLoo commented Mar 15, 2024

90566cbe2ca20c30bd953d6d245cd4d
b5133ae1af2b8269265b40899291c14
Hello, I have adjusted the parameters and tried many times, but the effect is still not very good. What is to be done?

@hongsukchoi
Copy link
Owner

hongsukchoi commented Mar 15, 2024 via email

@MilkLoo
Copy link
Author

MilkLoo commented Mar 16, 2024

I have not modified the code, is directly configured after the running code

@hongsukchoi
Copy link
Owner

hongsukchoi commented Mar 16, 2024 via email

@MilkLoo
Copy link
Author

MilkLoo commented Mar 18, 2024

985de15b5201cf3c8524f3df57a5737
57fc7fcd34c517b05d01a9172749753
Hello, I changed your parameter to True, but the effect is still the same.

@hongsukchoi
Copy link
Owner

Hi! It seems the code is giving random results. But I can't find a reason if you didn't change the code... I hope others people can try the current code and discuss. Unfortunately, I have no resource to run the code. Maybe you can check the inference code on 3DPW benchmark first.

@MilkLoo
Copy link
Author

MilkLoo commented Mar 22, 2024

Hello, I have solved this problem with the network after retraining, thank you!

@hongsukchoi
Copy link
Owner

@MilkLoo That sounds great! Maybe there's problem in the uploaded weight? I will check it out once I get resource again.

@MilkLoo
Copy link
Author

MilkLoo commented Apr 11, 2024

@hongsukchoi

I'm really sorry, I still need to trouble you, now I have the 2D pose data of Human3.6M test set, I want to use the model to evaluate the results, but the code of the evaluation part is not good at writing, can you share it? Thank you.

@hongsukchoi
Copy link
Owner

Hi!

I am not sure I understood your question. You could change the testing dataset here: https://github.com/hongsukchoi/3DCrowdNet_RELEASE/blob/main/assets/yaml/3dpw.yml

and run the command here:
https://github.com/hongsukchoi/3DCrowdNet_RELEASE/blob/main/assets/running.md

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants