How can I get one person date,not multi？

[GYPgalaxy]

I found that the FPS is just 1 ot 2, so I want to get one person date only. Maybe it will increase FPS？
If it doesn't work, what should I do to increase?

[Hzzone]

Decreasing the input image size may reduce the computational cost.

[GYPgalaxy]

I will try that.THX. :)

[GYPgalaxy]

By the way, I want to use the dataset pose_iter_584000.caffemodel , but I have no idea to use in torch . I'd love to know how you did it？！ >_<

[Hzzone]

This repo was done two years ago. If I remember correctly, I have used https://github.com/vadimkantorov/caffemodel2pytorch to convert the Caffe model to PyTorch, and then manually check the state_dict to match the PyTorch module. In other words, you first convert the weights, second, define your own PyTorch network like me, maybe you can visualize the Caffe network using https://ethereon.github.io/netscope/quickstart.html, and finally match the PyTorch module and converted Caffe weights.

[GYPgalaxy]

THX sooooo much !

[GYPgalaxy]

Sorry to bother you again. I want to know if your two variables candidate, subset have corresponding numbers with human key points? For example, what is the correspondence between the ID of each joint point officially pointed out and your two variables? Because I need to know which joint point corresponds to each coordinate.

[GYPgalaxy]

I found that even if the obtained joint points are not in order, your return data is arranged in order. So i'm confused.

[JonwayLuo]

Sorry to bother you again. I want to know if your two variables candidate, subset have corresponding numbers with human key points? For example, what is the correspondence between the ID of each joint point officially pointed out and your two variables? Because I need to know which joint point corresponds to each coordinate.

Hi, I have met the same question. I found that the keypoints return their shape as (18,4), but sometime it turns to (19,4) or (16,4) or less, sometimes even turns to (0,). I tried to use the Openpose's COCO18 format output as the input of D3DP, a 3D pose estimation method , but failed because of the changing outputs shape.