search

mks0601 / 3DMPPE_POSENET_RELEASE

Official PyTorch implementation of "Camera Distance-aware Top-down Approach for 3D Multi-person Pose Estimation from a Single RGB Image", ICCV 2019
MIT License
817 stars 147 forks source link

About inference my own images #132

Open Kazusaisjustice opened 1 year ago

Kazusaisjustice commented 1 year ago

Hello! Thanks for your great working! I'm confused about how to inference my own images, since when I run demo.py and change the input image to my image, the output seems wrong. Do you have any tutorial? In addition, if I want to inference a video, should I extract all frames of a video and then inference them one by one in your program?

Waiting for your reply! Thanks a lot!

mks0601 commented 1 year ago

You can refer to this: https://github.com/mks0601/3DMPPE_POSENET_RELEASE#quick-demo Yes. you should extract all frames.

Kazusaisjustice commented 1 year ago

Thanks for your reply. Actually I tried the quick-demo you mentioned, but when I changed the input image you supplied to my own image, the output 3d keypoints were wrong. The skeleton and the person did not overlap at all. Is this because the corresponding bbox and root depth are required to input each picture？Does that mean if I want to inference a video, I have to get the bbox and root depth of all frames?

mks0601 commented 1 year ago

As described in that link, you should set bbox and root depth for each image.

Kazusaisjustice commented 1 year ago

Well, I got it! Thanks a lot!