search

ankanbhunia / PIDM

Person Image Synthesis via Denoising Diffusion Model (CVPR 2023)
https://ankanbhunia.github.io/PIDM
MIT License
481 stars 62 forks source link

keypoints to .npy #38

Open jasonwongw opened 1 year ago

jasonwongw commented 1 year ago

Hello author, I got the key point of 18*2, using get_label_tensor function and using np.load to convert to npy file. When the last predict, the effect is poor and there is no image of the key points.

jasonwongw commented 1 year ago

out(1)(1)

ankanbhunia commented 1 year ago

Its clear that the key-points are not generated properly.

Can you provide more details ie. input and ouput of get_label_tensor in your case?

jasonwongw commented 1 year ago

The above problem, I have solved. However, when I use a .txt file with 18 key points generated by myself, the effect is poor. Here are my .txt files . 1.txt

Here is the rendering，I've iterated 100 times, but it's still not working. out

@ankanbhunia

jasonwongw commented 1 year ago

This is another frame of keypoint plots. out

jasonwongw commented 1 year ago

Hello author, I would like to ask, why is the face and costume different every time it is rendered?

chenbolin-master commented 1 year ago

@jasonwongw @ankanbhunia how to generate the reference_pose_0.npy from keypoints?

chenbolin-master commented 1 year ago

@jasonwongw how you do it ?(When the last predict, the effect is poor and there is no image of the key points.)