search

zju3dv / neuralbody

Code for "Neural Body: Implicit Neural Representations with Structured Latent Codes for Novel View Synthesis of Dynamic Humans" CVPR 2021 best paper candidate
Other
907 stars 130 forks source link

Mask generation like Zju-mocap dataset #142

Closed Dipankar1997161 closed 1 year ago

Dipankar1997161 commented 1 year ago

Hey @pengsida,

I would like to know, how were the "Mask_cihp" and "mask" generated for the respective images in zju-mocap dataset.

I have to create such masks for my custom dataset. Any specific model or software used in case of zju-mocap??

I have the 2d and 3d keypoints with me for my dataset, just require the masks.

I will wait for the response.

leviome commented 1 year ago

@Dipankar1997161 Hey bro, have you settled it down?

pengsida commented 1 year ago

You could use this repository to generate the masks: https://github.com/dendenxu/Self-Correction-Human-Parsing

treesturn commented 10 months ago

@Dipankar1997161 Hi can i ask how did u obtain the 2d and 3d keypoints for ur own dataset?

Dipankar1997161 commented 10 months ago

@Dipankar1997161 Hi can i ask how did u obtain the 2d and 3d keypoints for ur own dataset?

Hello there, So, for getting 2d joints you can either manually annotate it using CVAT or LabelMe tool.
Another way is to use pose-estimator (Openpose, yolo-HrNet) (you can find them in Easymocap GitHub which is linked here.)

Once you get the 2d joints, you can run the Mocap.py or the mv1p.py(multi_view) files under Easymocap GitHub and you will get the smpl as well as the 3d joints (if you have a monocular data, check it as well). Else you can use any 2d-3d pose_lift methods available.

treesturn commented 10 months ago

@Dipankar1997161 thanks for the suggestions, will try them out :)