search

karfly / learnable-triangulation-pytorch

This repository is an official PyTorch implementation of the paper "Learnable Triangulation of Human Pose" (ICCV 2019, oral). Proposed method archives state-of-the-art results in multi-view 3D human pose estimation!
MIT License
1.09k stars 181 forks source link

How does the number of camera views influence the performance on Human 3.6m dataset? #95

Closed huagl closed 3 years ago

huagl commented 4 years ago

Hi~ This repo and the methods that you proposed in your paper are very impressive! I wonder what's the performance when different number of camera views is used. May you report the experimental results?

pablovela5620 commented 4 years ago

In the CMU dataset section of the paper they had this chart showing performance of both the ransac and volumetric with the number of cameras image

huagl commented 4 years ago

In the CMU dataset section of the paper they had this chart showing performance of both the ransac and volumetric with the number of cameras image

Wow! Thank you very much. Actually, I want to know the performance with different number of camera views on Human 3.6m dataset. I didn't clarify my question clearly. Thanks anyway!

karfly commented 4 years ago

@huagl, on Human3.6M you can only measure performance for 2, 3 or 4 camera views.

huagl commented 4 years ago

@huagl, on Human3.6M you can only measure performance for 2, 3 or 4 camera views.

@karfly Yes, I know. I have a little problem to implement your codes on my machine, for the reason of the space of GPU. How about the performance when only using 2 or 3 camera views to train the network? Could you report your results when using different number of camera views on Human 3.6M? Thanks a lot.

karfly commented 4 years ago

@huagl, sorry, now I'm working on a different project, so I won't be able to calculate results for 2-, 3-view setup.

huagl commented 4 years ago

@huagl, sorry, now I'm working on a different project, so I won't be able to calculate results for 2-, 3-view setup.

@karfly That's all right. I will keep this issue open to see if anyone could help me. Thanks~